diff --git a/.DS_Store b/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..3cd874321e6200900eceb82183b38d047722f083
Binary files /dev/null and b/.DS_Store differ
diff --git a/data/.DS_Store b/data/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..fe0053e898f166dd4b77493fd20b302b63001a2b
Binary files /dev/null and b/data/.DS_Store differ
diff --git a/data/sampled_jsons/0.95_0.15_Contrastive_CRL_MCC_synthetic_real_ablation.jsonl b/data/sampled_jsons/0.95_0.15_Contrastive_CRL_MCC_synthetic_real_ablation.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..5328a1054bdd8c2cf869201cdb33a77e75b947bb
--- /dev/null
+++ b/data/sampled_jsons/0.95_0.15_Contrastive_CRL_MCC_synthetic_real_ablation.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "identification of nonparametric dynamic causal structure ...", "date": "", "ddg_snippet": "by M Fu · 2025 — Our framework is designed to address real -world climate scenarios characterized by general nonlinear causal relationships and latent variables.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2501.12500", "content": "by M Fu · 2025 — Our framework is designed to address real -world climate scenarios characterized by general nonlinear causal relationships and latent variables."}
+{"idx": 1, "title": "colorectal cancer", "date": "", "ddg_snippet": "19 Jan 2020 — real -world patients (RWP) treated with FOLFIXIRI 6 bevacizumab: A population-based retrospective cohort study. Tayyaba Bhatti, Michael Moser ...", "subpage_snippet": "", "source": "s3.amazonaws.com", "link": "https://s3.amazonaws.com/files.oncologymeetings.org/prod/s3fs-public/2020-01/GI20-COLORECTAL-CANCER.pdf?null", "content": "19 Jan 2020 — real -world patients (RWP) treated with FOLFIXIRI 6 bevacizumab: A population-based retrospective cohort study. Tayyaba Bhatti, Michael Moser ..."}
+{"idx": 2, "title": "TrustNLP 2023 The Third Workshop on Trustworthy Natural ...", "date": "", "ddg_snippet": "14 Jul 2023 — ... Ablation Study. We perform ablation studies to explore the effect of robustness improvement and explanation guided training for faithfulness ...", "subpage_snippet": "", "source": "aclanthology.org", "link": "https://aclanthology.org/2023.trustnlp-1.pdf", "content": "14 Jul 2023 — ... Ablation Study. We perform ablation studies to explore the effect of robustness improvement and explanation guided training for faithfulness ..."}
+{"idx": 3, "title": "Impurities and defects in, and isotope compositions of, ...", "date": "", "ddg_snippet": "Using the same offline laser ablation technique that was applied to the diamonds, followed by thermal ionization mass spectrometry (TIMS) we were able to ...", "subpage_snippet": "", "source": "scispace.com", "link": "https://scispace.com/pdf/impurities-and-defects-in-and-isotope-compositions-of-1kt50j4rru.pdf", "content": "Using the same offline laser ablation technique that was applied to the diamonds, followed by thermal ionization mass spectrometry (TIMS) we were able to ..."}
+{"idx": 4, "title": "Sunday, April 27, 2025", "date": "", "ddg_snippet": "27 Apr 2025 — This study investigates the real -world rates of genetics referrals in eligible cancer patients at Singapore's largest healthcare cluster using ... 6,076 pages", "subpage_snippet": "", "source": "www.aacr.org", "link": "https://www.aacr.org/wp-content/uploads/2025/05/AACR2025_Proceedings_050725.pdf", "content": "27 Apr 2025 — This study investigates the real -world rates of genetics referrals in eligible cancer patients at Singapore's largest healthcare cluster using ... 6,076 pages"}
+{"idx": 5, "title": "Autophagy-Dependent Generation of Free Fatty Acids Is ...", "date": "", "ddg_snippet": "by T Riffelmacher · 2017 · Cited by 316 — Autophagy-dependent generation of free fatty acids is critical for normal neutrophil differentiation.", "subpage_snippet": "", "source": "pmc.ncbi.nlm.nih.gov", "link": "https://pmc.ncbi.nlm.nih.gov/articles/PMC5610174/", "content": "by T Riffelmacher · 2017 · Cited by 316 — Autophagy-dependent generation of free fatty acids is critical for normal neutrophil differentiation."}
+{"idx": 6, "title": "Public Health Goals for Haloacetic Acids - OEHHA", "date": "", "ddg_snippet": "16 Dec 2022 — Contributors. Christopher Banks, Ph.D. Joseph Brown, Ph.D. John Budroe, Ph.D. Vincent Cogliano, Ph.D. James Donald, Ph.D. 499 pages", "subpage_snippet": "", "source": "oehha.ca.gov", "link": "https://oehha.ca.gov/sites/default/files/media/downloads/water/chemicals/phg/haaphg123022.pdf", "content": "16 Dec 2022 — Contributors. Christopher Banks, Ph.D. Joseph Brown, Ph.D. John Budroe, Ph.D. Vincent Cogliano, Ph.D. James Donald, Ph.D. 499 pages"}
+{"idx": 7, "title": "Recent Advances of Novel Pharmaceutical Designs for Anti ...", "date": "", "ddg_snippet": "0.95 ± 0.15 . 5a. 3.9 ± 0.71. 1.5 ± 0.17. 5b. >40. 29.8 ± 0.52. 5c. >40. 27.33 ± 0.83. 5d. >40. 29.2 ± 0.52. 5e. >40. 28.13 ± 0.83. 5f. >40. 4.5 ± 0.28. 5g. 23.2 ...", "subpage_snippet": "", "source": "mdpi-res.com", "link": "https://mdpi-res.com/bookfiles/book/7693/Recent_Advances_of_Novel_Pharmaceutical_Designs_for_Anticancer_Therapies.pdf?v=1718125915", "content": "0.95 ± 0.15 . 5a. 3.9 ± 0.71. 1.5 ± 0.17. 5b. >40. 29.8 ± 0.52. 5c. >40. 27.33 ± 0.83. 5d. >40. 29.2 ± 0.52. 5e. >40. 28.13 ± 0.83. 5f. >40. 4.5 ± 0.28. 5g. 23.2 ..."}
+{"idx": 8, "title": "Mps1 Regulates Kinetochore-Microtubule Attachment Stability ...", "date": "", "ddg_snippet": "The spindle assembly checkpoint kinase Mps1 not only inhibits anaphase but also corrects erro- neous attachments that could lead to missegregation.", "subpage_snippet": "", "source": "www.cell.com", "link": "https://www.cell.com/cms/10.1016/j.devcel.2017.03.025/attachment/a67cfe70-dc22-4ab1-ac64-4b30d86f3452/mmc3.pdf", "content": "The spindle assembly checkpoint kinase Mps1 not only inhibits anaphase but also corrects erro- neous attachments that could lead to missegregation."}
+{"idx": 9, "title": "Capillary function in patients with chronic venous insufficiency", "date": "", "ddg_snippet": "It has been demonstrated in normal subjects that persistently raised venous pressure results in trapping of leucocytes in the peripheral circulation'.", "subpage_snippet": "", "source": "academic.oup.com", "link": "https://academic.oup.com/bjs/article-pdf/75/6/597/59386982/bjs1800750635.pdf", "content": "It has been demonstrated in normal subjects that persistently raised venous pressure results in trapping of leucocytes in the peripheral circulation'."}
diff --git a/data/sampled_jsons/7uqVfZW6Mo_Not_All_Diffusion_Model_Activations_Have_Been_Evaluated_as_Discriminative_Features_Sectio.jsonl b/data/sampled_jsons/7uqVfZW6Mo_Not_All_Diffusion_Model_Activations_Have_Been_Evaluated_as_Discriminative_Features_Sectio.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..6fe8dcc531f042547dfd60d420278e68a658486d
--- /dev/null
+++ b/data/sampled_jsons/7uqVfZW6Mo_Not_All_Diffusion_Model_Activations_Have_Been_Evaluated_as_Discriminative_Features_Sectio.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Not All Diffusion Model Activations Have Been Evaluated as ...", "date": "", "ddg_snippet": "Diffusion models are initially designed for image generation. Recent research shows that the internal signals within their backbones, named activations , can also serve as dense features for various discriminative tasks such as semantic segmentation. Given numerous activations , selecting a small yet effective subset poses a fundamental problem. To this end, the early study of this field ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2410.03558", "content": "Diffusion models are initially designed for image generation. Recent research shows that the internal signals within their backbones, named activations , can also serve as dense features for various discriminative tasks such as semantic segmentation. Given numerous activations , selecting a small yet effective subset poses a fundamental problem. To this end, the early study of this field ..."}
+{"idx": 1, "title": "Darkbblue/generic-diffusion-feature - GitHub", "date": "", "ddg_snippet": "Diffusion feature is a quite popular way to utilize generative diffusion models for discrimination. It's very simple: just extract some internal activations from a diffusion model , and then use these 2D features to replace image inputs of any discriminative model . There have been quite many diffusion feature studies. But we notice that almost all of them experiment with Stable Diffusion v1.4 ...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/Darkbblue/generic-diffusion-feature", "content": "Diffusion feature is a quite popular way to utilize generative diffusion models for discrimination. It's very simple: just extract some internal activations from a diffusion model , and then use these 2D features to replace image inputs of any discriminative model . There have been quite many diffusion feature studies. But we notice that almost all of them experiment with Stable Diffusion v1.4 ..."}
+{"idx": 2, "title": "Not All Diffusion Model Activations Have Been Evaluated as ...", "date": "", "ddg_snippet": "To this end, the early study of this field performs a large-scale quantitative comparison of the discriminative ability of the activations . However, we find that many potential activations have not been evaluated , such as the queries and keys used to compute attention scores.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=7uqVfZW6Mo", "content": "To this end, the early study of this field performs a large-scale quantitative comparison of the discriminative ability of the activations . However, we find that many potential activations have not been evaluated , such as the queries and keys used to compute attention scores."}
+{"idx": 3, "title": "Not All Diffusion Model Activations Have Been Evaluated as ...", "date": "", "ddg_snippet": "However, we find that many potential activations have not been evaluated , such as the queries and keys used to compute attention scores. Moreover, recent advancements in diffusion architectures bring many new activations , such as those within embedded ViT modules. Both combined, activation selection remains unresolved but overlooked.", "subpage_snippet": "", "source": "proceedings.neurips.cc", "link": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/633780c1344d0c95e4d2dd3431fe08d9-Abstract-Conference.html", "content": "However, we find that many potential activations have not been evaluated , such as the queries and keys used to compute attention scores. Moreover, recent advancements in diffusion architectures bring many new activations , such as those within embedded ViT modules. Both combined, activation selection remains unresolved but overlooked."}
+{"idx": 4, "title": "Not All Diffusion Model Activations Have Been Evaluated as ...", "date": "", "ddg_snippet": "Unlocking superior discriminative features from diffusion models , this research reveals key activation properties for effective feature selection, surpassing state-of-the-art methods.", "subpage_snippet": "", "source": "deep-diver.github.io", "link": "https://deep-diver.github.io/neurips2024/spotlight-others/7uqvfzw6mo/", "content": "Unlocking superior discriminative features from diffusion models , this research reveals key activation properties for effective feature selection, surpassing state-of-the-art methods."}
+{"idx": 5, "title": "PDF Not All Diffusion Model Activations Have Been Evaluated as ...", "date": "", "ddg_snippet": "Not All Diffusion Model Activations Have Been Evaluated as Discriminative Features Benyuan Meng, Qianqian Xu*, Zitai Wang, Xiaochun Cao, Qingming Huang*", "subpage_snippet": "", "source": "nips.cc", "link": "https://nips.cc/media/neurips-2024/Slides/96411.pdf", "content": "Not All Diffusion Model Activations Have Been Evaluated as Discriminative Features Benyuan Meng, Qianqian Xu*, Zitai Wang, Xiaochun Cao, Qingming Huang*"}
+{"idx": 6, "title": "Not All Diffusion Model Activations Have Been Evaluated as ...", "date": "", "ddg_snippet": "View recent discussion. Abstract: Diffusion models are initially designed for image generation. Recent research shows that the internal signals within their backbones, named activations , can also serve as dense features for various discriminative tasks such as semantic segmentation. Given numerous activations , selecting a small yet effective subset poses a fundamental problem. To this end, the ...", "subpage_snippet": "", "source": "www.alphaxiv.org", "link": "https://www.alphaxiv.org/overview/2410.03558v3", "content": "View recent discussion. Abstract: Diffusion models are initially designed for image generation. Recent research shows that the internal signals within their backbones, named activations , can also serve as dense features for various discriminative tasks such as semantic segmentation. Given numerous activations , selecting a small yet effective subset poses a fundamental problem. To this end, the ..."}
+{"idx": 7, "title": "Not All Diffusion Model Activations Have Been Evaluated as ...", "date": "", "ddg_snippet": "Through a series of experiments, the paper shows that not all diffusion model activations are equally useful as features for image classification. This suggests that there may be untapped potential in these internal representations that could be unlocked through further research.", "subpage_snippet": "", "source": "www.aimodels.fyi", "link": "https://www.aimodels.fyi/papers/arxiv/not-all-diffusion-model-activations-have-been", "content": "Through a series of experiments, the paper shows that not all diffusion model activations are equally useful as features for image classification. This suggests that there may be untapped potential in these internal representations that could be unlocked through further research."}
+{"idx": 8, "title": "Not All Diffusion Model Activations Have Been Evaluated as ...", "date": "", "ddg_snippet": "Table 2: Experimental results on semantic segmentation and its altered version with scarce labeled data, evaluated using mIoU↑ metric. The best results are in bold font and the runner-up is underlined. - \"Not All Diffusion Model Activations Have Been Evaluated as Discriminative Features \"", "subpage_snippet": "", "source": "www.semanticscholar.org", "link": "https://www.semanticscholar.org/paper/Not-All-Diffusion-Model-Activations-Have-Been-as-Meng-Xu/6fce3070fbbeb8b8eb86e270ec341c155e941905/figure/3", "content": "Table 2: Experimental results on semantic segmentation and its altered version with scarce labeled data, evaluated using mIoU↑ metric. The best results are in bold font and the runner-up is underlined. - \"Not All Diffusion Model Activations Have Been Evaluated as Discriminative Features \""}
+{"idx": 9, "title": "CatnissCullen/generic-diffusion-feature_forked - GitHub", "date": "", "ddg_snippet": "Diffusion feature is a quite popular way to utilize generative diffusion models for discrimination. It's very simple: just extract some internal activations from a diffusion model , and then use these 2D features to replace image inputs of any discriminative model . There have been quite many diffusion feature studies.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/CatnissCullen/generic-diffusion-feature_forked", "content": "Diffusion feature is a quite popular way to utilize generative diffusion models for discrimination. It's very simple: just extract some internal activations from a diffusion model , and then use these 2D features to replace image inputs of any discriminative model . There have been quite many diffusion feature studies."}
diff --git a/data/sampled_jsons/ACIDDnTbSJ_Feint_Behaviors_Strategies_scheduler_weights_lambda_short_lambda_long_Section_4.2.2.jsonl b/data/sampled_jsons/ACIDDnTbSJ_Feint_Behaviors_Strategies_scheduler_weights_lambda_short_lambda_long_Section_4.2.2.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..07331073c045faf6fa1bb1569cdda84a644e0c1d
--- /dev/null
+++ b/data/sampled_jsons/ACIDDnTbSJ_Feint_Behaviors_Strategies_scheduler_weights_lambda_short_lambda_long_Section_4.2.2.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Feint Behaviors and Strategies : Formalization, Implementation and...", "date": "", "ddg_snippet": "Full paper. ACIDDnTbSJ . Junyu Liu et el.Ultimately, this section aims to demonstrate the feasibility and practicality of incorporating Feint behaviors into real-world MARL applications, highlighting its potential for enhancing the performance and strategic depth of multi-agent systems.", "subpage_snippet": "", "source": "deep-diver.github.io", "link": "https://deep-diver.github.io/neurips2024/posters/aciddntbsj/", "content": "Full paper. ACIDDnTbSJ . Junyu Liu et el.Ultimately, this section aims to demonstrate the feasibility and practicality of incorporating Feint behaviors into real-world MARL applications, highlighting its potential for enhancing the performance and strategic depth of multi-agent systems."}
+{"idx": 1, "title": "Feint Behaviors and Strategies : Formalization, Implementation and...", "date": "", "ddg_snippet": "Feint behaviors refer to a set of deceptive behaviors in a nuanced manner, which enable players to obtain temporal and spatial advantages over opponents in competitive games.", "subpage_snippet": "", "source": "papers.cool", "link": "https://papers.cool/venue/ACIDDnTbSJ@OpenReview", "content": "Feint behaviors refer to a set of deceptive behaviors in a nuanced manner, which enable players to obtain temporal and spatial advantages over opponents in competitive games."}
+{"idx": 2, "title": "差分注意力，负注意力的引入-CSDN博客", "date": "", "ddg_snippet": "self. lambda _q1 = nn.Parameter(torch.zeros(self.head_dim, dtype=torch.float32).normal_(mean=0,std=0.1)).attn_ weights = attn_ weights .view(bsz, self.num_heads, 2, tgt_len, src_len) #[131072, 8, 2 , 2 , 2 ] 第一个2是两个差分.", "subpage_snippet": "", "source": "blog.csdn.net", "link": "https://blog.csdn.net/weixin_45668967/article/details/142865222", "content": "self. lambda _q1 = nn.Parameter(torch.zeros(self.head_dim, dtype=torch.float32).normal_(mean=0,std=0.1)).attn_ weights = attn_ weights .view(bsz, self.num_heads, 2, tgt_len, src_len) #[131072, 8, 2 , 2 , 2 ] 第一个2是两个差分."}
+{"idx": 3, "title": "Lambda в Python: синтаксис, аргументы и много примеров...", "date": "", "ddg_snippet": "Все, что нужно знать о lambda -функциях в Python. В этой статье вы узнаете о том, что такое лямбда-функции в Python.", "subpage_snippet": "", "source": "PythonRu.com", "link": "https://PythonRu.com/osnovy/vse-chto-nuzhno-znat-o-lambda-funkcijah-v-python", "content": "Все, что нужно знать о lambda -функциях в Python. В этой статье вы узнаете о том, что такое лямбда-функции в Python."}
+{"idx": 4, "title": "How to change lambda sensor on VW GOLF 4 [TUTORIAL...] - YouTube", "date": "", "ddg_snippet": "How to change lambda sensor / oxygen sensor / o2 sensor on VOLKSWAGEN GOLF 4 (1J1) 1.6 Hatchback 1997–2005 [TUTORIAL AUTODOC] Lambda sensor on VW Bora Saloon ...", "subpage_snippet": "", "source": "www.youtube.com", "link": "https://www.youtube.com/watch?v=zgqYEh7SDo4", "content": "How to change lambda sensor / oxygen sensor / o2 sensor on VOLKSWAGEN GOLF 4 (1J1) 1.6 Hatchback 1997–2005 [TUTORIAL AUTODOC] Lambda sensor on VW Bora Saloon ..."}
+{"idx": 5, "title": "c++11 - C++ shorter lambda syntax - Stack Overflow", "date": "", "ddg_snippet": "Some of the columns require longer lambdas , but as it is the syntax for writing the lambda is about as long as the content it self. Can the lambda syntax be reduced at all? (say by implicit argument or implicitly return the last statement).", "subpage_snippet": "", "source": "stackoverflow.com", "link": "https://stackoverflow.com/questions/36775231/c-shorter-lambda-syntax", "content": "Some of the columns require longer lambdas , but as it is the syntax for writing the lambda is about as long as the content it self. Can the lambda syntax be reduced at all? (say by implicit argument or implicitly return the last statement)."}
+{"idx": 6, "title": "Sci-Hub: наука и коммунизм — неотделимы!", "date": "", "ddg_snippet": "3.0.co;2-3\",\"title\":\"Predominant expression of lambda light chain in adult cases with non-T-cell acute lymphocytic and chronic myelogenous leukemia in lymphoid blast crisis\",\"journal\":\"Cancer\",\"author\":\"Mori\",\"year\":1991,\"country\":\"IN\"}\">11:33:01 /", "subpage_snippet": "", "source": "www.sci-hub.red", "link": "https://www.sci-hub.red/", "content": "3.0.co;2-3\",\"title\":\"Predominant expression of lambda light chain in adult cases with non-T-cell acute lymphocytic and chronic myelogenous leukemia in lymphoid blast crisis\",\"journal\":\"Cancer\",\"author\":\"Mori\",\"year\":1991,\"country\":\"IN\"}\">11:33:01 /"}
+{"idx": 7, "title": "Your Guide to Foundry Freaks in Borderlands 4: Location, Drops, How...", "date": "", "ddg_snippet": "Suggested strategy . Frank the Furnace.Step 5. After defeating all three, immediately check loot drops and extract quickly, as respawn timers are short and more mobs may appear for repeat runs. Why Foundry Freaks are worth farming.", "subpage_snippet": "", "source": "nerdschalk.com", "link": "https://nerdschalk.com/your-guide-to-foundry-freaks-in-borderlands-4-location-drops-how-to-farm/", "content": "Suggested strategy . Frank the Furnace.Step 5. After defeating all three, immediately check loot drops and extract quickly, as respawn timers are short and more mobs may appear for repeat runs. Why Foundry Freaks are worth farming."}
+{"idx": 8, "title": "Lambda expressions (since C++11) - cppreference.com", "date": "", "ddg_snippet": "Metaprogramming library. General utilities library. Containers library. Iterators library. Ranges library. Algorithms library. Strings library. Text processing library. Numerics library. Date and time library. Input/output library. Filesystem library...", "subpage_snippet": "", "source": "en.cppreference.com", "link": "https://en.cppreference.com/w/cpp/language/lambda.html", "content": "Metaprogramming library. General utilities library. Containers library. Iterators library. Ranges library. Algorithms library. Strings library. Text processing library. Numerics library. Date and time library. Input/output library. Filesystem library..."}
+{"idx": 9, "title": "Iuno Best Builds and Teams | Wuthering Waves｜Game8", "date": "", "ddg_snippet": "2.7 Livestream scheduled for 9/26! Version 2.6 Phase 2 is out now! Iuno Debut: Best Build - Materials - Weapon Rerun: Ciaccona Build - Materials Area: Sanguis Plateaus - 2.6 Sonance Caskets Events: Tidal Defense Sim, Marks of the Wild.", "subpage_snippet": "", "source": "game8.co", "link": "https://game8.co/games/Wuthering-Waves/archives/524889", "content": "2.7 Livestream scheduled for 9/26! Version 2.6 Phase 2 is out now! Iuno Debut: Best Build - Materials - Weapon Rerun: Ciaccona Build - Materials Area: Sanguis Plateaus - 2.6 Sonance Caskets Events: Tidal Defense Sim, Marks of the Wild."}
diff --git a/data/sampled_jsons/A_Geometric_Approach_to_Personalized_Recommendation_Set-Theoretic_Constraints_Box_Embeddings_Table_1.jsonl b/data/sampled_jsons/A_Geometric_Approach_to_Personalized_Recommendation_Set-Theoretic_Constraints_Box_Embeddings_Table_1.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..c64b609ebb66716a7870b44f4f9de35bc77dd5e4
--- /dev/null
+++ b/data/sampled_jsons/A_Geometric_Approach_to_Personalized_Recommendation_Set-Theoretic_Constraints_Box_Embeddings_Table_1.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "A Geometric Approach to Personalized Recommendation with...", "date": "", "ddg_snippet": "22 Jun 2025 — In this work, we formulate the problem of personalized item recommendation as matrix completion where rows are set -theoretically dependent. ... Queries involving ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=27tMzmzDjO", "content": "22 Jun 2025 — In this work, we formulate the problem of personalized item recommendation as matrix completion where rows are set -theoretically dependent. ... Queries involving ..."}
+{"idx": 1, "title": "A Geometric Approach to Personalized Recommendation ...", "date": "", "ddg_snippet": "15 Feb 2025 — We empirically demonstrate the superiority of box embeddings over vector-based neural methods on both simple and complex item recommendation ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2502.10875v1", "content": "15 Feb 2025 — We empirically demonstrate the superiority of box embeddings over vector-based neural methods on both simple and complex item recommendation ..."}
+{"idx": 2, "title": "A Geometric Approach to Personalized Recommendation ...", "date": "", "ddg_snippet": "by S Dasgupta · 2025 — Box embeddings, with their geometric set operations, sig- nificantly outperform all vector-based methods.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2502.10875", "content": "by S Dasgupta · 2025 — Box embeddings, with their geometric set operations, sig- nificantly outperform all vector-based methods."}
+{"idx": 3, "title": "[Literature Review] A Geometric Approach to Personalized ...", "date": "", "ddg_snippet": "This page provides the most accurate and concise summary worldwide for the paper titled A Geometric Approach to Personalized Recommendation with Set - Theoretic ...", "subpage_snippet": "", "source": "www.themoonlight.io", "link": "https://www.themoonlight.io/en/review/a-geometric-approach-to-personalized-recommendation-with-set-theoretic-constraints-using-box-embeddings", "content": "This page provides the most accurate and concise summary worldwide for the paper titled A Geometric Approach to Personalized Recommendation with Set - Theoretic ..."}
+{"idx": 4, "title": "A Geometric Approach to Personalized Recommendation ...", "date": "", "ddg_snippet": "We empirically demonstrate the superiority of box embeddings over vector-based neural methods on both simple and complex item recommendation queries by up to 30 ...", "subpage_snippet": "", "source": "chatpaper.com", "link": "https://chatpaper.com/paper/165898", "content": "We empirically demonstrate the superiority of box embeddings over vector-based neural methods on both simple and complex item recommendation queries by up to 30 ..."}
+{"idx": 5, "title": "Lyz103/Recommendation-paper-daily", "date": "", "ddg_snippet": "A Geometric Approach to Personalized Recommendation with Set - Theoretic Constraints Using Box Embeddings ... Data Set Terminology of Artificial Intelligence ...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/Lyz103/Recommendation-paper-daily", "content": "A Geometric Approach to Personalized Recommendation with Set - Theoretic Constraints Using Box Embeddings ... Data Set Terminology of Artificial Intelligence ..."}
+{"idx": 6, "title": "InBox: Recommendation with Knowledge Graph using ...", "date": "", "ddg_snippet": "by Z Xu · Cited by 1 — Table 1 : Statistics of the datasets used in our experiments. We report the ... Theoretic Semantics of Words using Box Embeddings . In Proceedings of the ...", "subpage_snippet": "", "source": "www.vldb.org", "link": "https://www.vldb.org/pvldb/vol17/p4641-xu.pdf", "content": "by Z Xu · Cited by 1 — Table 1 : Statistics of the datasets used in our experiments. We report the ... Theoretic Semantics of Words using Box Embeddings . In Proceedings of the ..."}
+{"idx": 7, "title": "Track: Poster Session 5 East", "date": "", "ddg_snippet": "17 Jul 2025 — A Geometric Approach to Personalized Recommendation with Set - Theoretic Constraints Using Box Embeddings ... data sparsity, which is most ...", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/session/50267", "content": "17 Jul 2025 — A Geometric Approach to Personalized Recommendation with Set - Theoretic Constraints Using Box Embeddings ... data sparsity, which is most ..."}
+{"idx": 8, "title": "GeOKG: geometry-aware knowledge graph embedding for ...", "date": "", "ddg_snippet": "by CU Jeong · 2025 — The EL Embeddings approach , for instance, constructs a geometric model that satisfies EL++ logic constraints , defining bespoke scoring ...", "subpage_snippet": "", "source": "academic.oup.com", "link": "https://academic.oup.com/bioinformatics/article/41/4/btaf160/8111648", "content": "by CU Jeong · 2025 — The EL Embeddings approach , for instance, constructs a geometric model that satisfies EL++ logic constraints , defining bespoke scoring ..."}
+{"idx": 9, "title": "Accurate Neural Reasoning for Question-Answering", "date": "", "ddg_snippet": "This work proposes here a novel KB embedding scheme that supports generalization, but also allows accurate logical reasoning with a KB, and introduces two ...", "subpage_snippet": "", "source": "www.semanticscholar.org", "link": "https://www.semanticscholar.org/paper/Guessing-What's-Plausible-But-Remembering-What's-Sun-Arnold/cd9e1eac4c93a314254cf8a8682ed5f01b6a808f/figure/3", "content": "This work proposes here a novel KB embedding scheme that supports generalization, but also allows accurate logical reasoning with a KB, and introduces two ..."}
diff --git a/data/sampled_jsons/Algorithm_2_k-HOC_time_complexity_O(_siteopenreview.net.jsonl b/data/sampled_jsons/Algorithm_2_k-HOC_time_complexity_O(_siteopenreview.net.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..d51ca270c443ae7733645bf5c214041bc8006590
--- /dev/null
+++ b/data/sampled_jsons/Algorithm_2_k-HOC_time_complexity_O(_siteopenreview.net.jsonl
@@ -0,0 +1,3 @@
+{"idx": 0, "title": "HIERARCHICAL OVERLAPPING CLUSTERING FUNCTION ALGORITHM AND ...", "date": "", "ddg_snippet": "051 (2) Approximation algorithm. Based on our cost function, we formulate the primal and the 052 dual versions of HOC, respectively. We provide an a = √ 2 − Θ(1+ε )-approximation algorithm 3 6 n 053 ( Algorithm 2 ) for the dual k-HOC problem, where k ∈ Z+ is an upper bound of key clusters (explained in Definition 2.10). Our algorithm is a recursive process of overlapping bipartition in ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=oHSXRy29tj", "content": "051 (2) Approximation algorithm. Based on our cost function, we formulate the primal and the 052 dual versions of HOC, respectively. We provide an a = √ 2 − Θ(1+ε )-approximation algorithm 3 6 n 053 ( Algorithm 2 ) for the dual k-HOC problem, where k ∈ Z+ is an upper bound of key clusters (explained in Definition 2.10). Our algorithm is a recursive process of overlapping bipartition in ..."}
+{"idx": 1, "title": "Hierarchical overlapping clustering: cost function, algorithm ...", "date": "", "ddg_snippet": "by Y Pan — A major contribution of Orecchia et al. 2022 is the practical nearly linear time complexity of their approximation algorithm. They use the seminal work of Chen ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=oHSXRy29tj", "content": "by Y Pan — A major contribution of Orecchia et al. 2022 is the practical nearly linear time complexity of their approximation algorithm. They use the seminal work of Chen ..."}
+{"idx": 2, "title": "Hierarchical Overlapping Clustering on Graphs: Cost Function, ...", "date": "", "ddg_snippet": ", 2019) to O (log n) and O (. √ log n), respectively. It ... Algorithm 2 Algorithm for k-HOC . Input: an ... The time complexity of Algorithm 1 is O (n4 log m.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/attachment?id=51x0dfsD8A&name=pdf", "content": ", 2019) to O (log n) and O (. √ log n), respectively. It ... Algorithm 2 Algorithm for k-HOC . Input: an ... The time complexity of Algorithm 1 is O (n4 log m."}
diff --git a/data/sampled_jsons/CVE-Bench_Success@5_T-Agent_one-day_setting_Figure_3_year_2024.jsonl b/data/sampled_jsons/CVE-Bench_Success@5_T-Agent_one-day_setting_Figure_3_year_2024.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..a1d57e18ae55e82d66d6106d8775fe2ab4807c6a
--- /dev/null
+++ b/data/sampled_jsons/CVE-Bench_Success@5_T-Agent_one-day_setting_Figure_3_year_2024.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "CVE-Bench: A Benchmark for AI Agents’ Ability to Exploit", "date": "", "ddg_snippet": "We apply CVE - Bench to evaluate various LLM agents under both zero- day and one - day settings . ... agent framework, teams of LLM agents (Fang et al., ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2503.17332v4", "content": "We apply CVE - Bench to evaluate various LLM agents under both zero- day and one - day settings . ... agent framework, teams of LLM agents (Fang et al., ..."}
+{"idx": 1, "title": "Establishing Best Practices for Building Rigorous Agentic", "date": "", "ddg_snippet": "To evaluate AI agents , researchers and practitioners have built agentic benchmarks with realistic tasks to track progress and assist decision-making [ ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2507.02825v5", "content": "To evaluate AI agents , researchers and practitioners have built agentic benchmarks with realistic tasks to track progress and assist decision-making [ ..."}
+{"idx": 2, "title": "ExCyTIn-Bench: Evaluating LLM agents on Cyber Threat", "date": "", "ddg_snippet": "... advancement of Large Language Models (LLMs) has enabled astonishing achievements in complex tasks [ 54 , 14 , 45 , 47 , 46 ] , that LLM agents can ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2507.14201v1", "content": "... advancement of Large Language Models (LLMs) has enabled astonishing achievements in complex tasks [ 54 , 14 , 45 , 47 , 46 ] , that LLM agents can ..."}
+{"idx": 3, "title": "Establishing Best Practices for Building Rigorous Agentic", "date": "", "ddg_snippet": "To evaluate AI agents , researchers and practitioners have built agentic benchmarks with realistic tasks to track progress and assist decision-making [ ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2507.02825v1", "content": "To evaluate AI agents , researchers and practitioners have built agentic benchmarks with realistic tasks to track progress and assist decision-making [ ..."}
+{"idx": 4, "title": "Establishing Best Practices for Building Rigorous Agentic", "date": "", "ddg_snippet": "To evaluate AI agents , researchers and practitioners have built agentic benchmarks with realistic tasks to track progress and assist decision-making [ ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2507.02825v4", "content": "To evaluate AI agents , researchers and practitioners have built agentic benchmarks with realistic tasks to track progress and assist decision-making [ ..."}
+{"idx": 5, "title": "Blog post — Zero Day Initiative — Blog", "date": "", "ddg_snippet": "The event garnered more participation ... That event awarded $ 1 ,132,500 for 29 unique 0- days and also saw the first Docker escape at a Pwn2Own event.", "subpage_snippet": "", "source": "www.thezdi.com", "link": "https://www.thezdi.com/blog/category/Blog+post", "content": "The event garnered more participation ... That event awarded $ 1 ,132,500 for 29 unique 0- days and also saw the first Docker escape at a Pwn2Own event."}
+{"idx": 6, "title": "How Does Time Horizon Vary Across Domains? - METR", "date": "", "ddg_snippet": "In video understanding (VideoMME, CGBench), models are capable of answering questions on ~ 1 hour videos with > 50% success rates, but time horizon ...", "subpage_snippet": "", "source": "evals.alignment.org", "link": "https://evals.alignment.org/blog/2025-07-14-how-does-time-horizon-vary-across-domains/", "content": "In video understanding (VideoMME, CGBench), models are capable of answering questions on ~ 1 hour videos with > 50% success rates, but time horizon ..."}
+{"idx": 7, "title": "AI-Driven Cyberattacks are on the Rise. Are You Ready? -", "date": "", "ddg_snippet": "Figure 1 : CVE - Bench framework architecture showing LLM ... Figure 3 : Success rates of LLM agents exploiting zero- day and one - day vulnerabilities.", "subpage_snippet": "", "source": "lsvp.com", "link": "https://lsvp.com/stories/ai-enabled-hacking-is-here-are-we-ready-for-it/", "content": "Figure 1 : CVE - Bench framework architecture showing LLM ... Figure 3 : Success rates of LLM agents exploiting zero- day and one - day vulnerabilities."}
+{"idx": 8, "title": "METR: How Does Time Horizon Vary Across Domains? - LessWrong", "date": "", "ddg_snippet": "In video understanding (VideoMME, CGBench), models are capable of answering questions on ~ 1 hour videos with > 50% success rates, but time horizon ...", "subpage_snippet": "", "source": "www.greaterwrong.com", "link": "https://www.greaterwrong.com/posts/6KcP7tEe5hgvHbrSF/metr-how-does-time-horizon-vary-across-domains", "content": "In video understanding (VideoMME, CGBench), models are capable of answering questions on ~ 1 hour videos with > 50% success rates, but time horizon ..."}
+{"idx": 9, "title": "How Does Time Horizon Vary Across Domains? - METR", "date": "", "ddg_snippet": "In video understanding (VideoMME, CGBench), models are capable of answering questions on ~ 1 hour videos with > 50% success rates, but time horizon ...", "subpage_snippet": "", "source": "metr.org", "link": "https://metr.org/blog/2025-07-14-how-does-time-horizon-vary-across-domains/", "content": "In video understanding (VideoMME, CGBench), models are capable of answering questions on ~ 1 hour videos with > 50% success rates, but time horizon ..."}
diff --git a/data/sampled_jsons/CVE-Bench_paper_Common_Failure_Modes_section_Table_5_Insufficient_Exploration_'Insufficient_Explorat_year_2023-2024.jsonl b/data/sampled_jsons/CVE-Bench_paper_Common_Failure_Modes_section_Table_5_Insufficient_Exploration_'Insufficient_Explorat_year_2023-2024.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..646844cf8f8c32a764573b4baf5c0f4f9310d6a6
--- /dev/null
+++ b/data/sampled_jsons/CVE-Bench_paper_Common_Failure_Modes_section_Table_5_Insufficient_Exploration_'Insufficient_Explorat_year_2023-2024.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Common Vulnerabilities and Exposures - Wikipedia", "date": "", "ddg_snippet": "Logo The Common Vulnerabilities and Exposures (CVE) system, originally Common Vulnerability Enumeration, [1] provides a reference method for publicly known information-security …", "subpage_snippet": "", "source": "en.wikipedia.org", "link": "https://en.wikipedia.org/wiki/Common_Vulnerabilities_and_Exposures", "content": "Logo The Common Vulnerabilities and Exposures (CVE) system, originally Common Vulnerability Enumeration, [1] provides a reference method for publicly known information-security …"}
+{"idx": 1, "title": "CVE : Common Vulnerabilities and Exposures", "date": "", "ddg_snippet": "At cve.org, we provide the authoritative reference method for publicly known information-security vulnerabilities and exposures", "subpage_snippet": "", "source": "www.cve.org", "link": "https://www.cve.org/", "content": "At cve.org, we provide the authoritative reference method for publicly known information-security vulnerabilities and exposures"}
+{"idx": 2, "title": "CVE security vulnerability database. Security vulnerabilities, …", "date": "", "ddg_snippet": "CVEDetails.com is a vulnerability intelligence solution providing CVE security vulnerability database, exploits, advisories, product and CVE risk scores, attack surface intelligence, open source …", "subpage_snippet": "", "source": "www.cvedetails.com", "link": "https://www.cvedetails.com/", "content": "CVEDetails.com is a vulnerability intelligence solution providing CVE security vulnerability database, exploits, advisories, product and CVE risk scores, attack surface intelligence, open source …"}
+{"idx": 3, "title": "What is a CVE ? - Red Hat", "date": "", "ddg_snippet": "Sep 4, 2024 · CVE, short for Common Vulnerabilities and Exposures, is a list of publicly disclosed computer security flaws.", "subpage_snippet": "", "source": "www.redhat.com", "link": "https://www.redhat.com/en/topics/security/what-is-cve", "content": "Sep 4, 2024 · CVE, short for Common Vulnerabilities and Exposures, is a list of publicly disclosed computer security flaws."}
+{"idx": 4, "title": "CVEs and Security Vulnerabilities - OpenCVE", "date": "", "ddg_snippet": "Explore the latest vulnerabilities and security issues in the CVE database", "subpage_snippet": "", "source": "app.opencve.io", "link": "https://app.opencve.io/", "content": "Explore the latest vulnerabilities and security issues in the CVE database"}
+{"idx": 5, "title": "What is CVE and CVSS | Vulnerability Scoring Explained | Imperva", "date": "", "ddg_snippet": "Jul 8, 2025 · What is the Common Vulnerabilities and Exposures (CVE) Glossary CVE stands for Common Vulnerabilities and Exposures. CVE is a glossary that classifies vulnerabilities. The …", "subpage_snippet": "", "source": "www.imperva.com", "link": "https://www.imperva.com/learn/application-security/cve-cvss-vulnerability/", "content": "Jul 8, 2025 · What is the Common Vulnerabilities and Exposures (CVE) Glossary CVE stands for Common Vulnerabilities and Exposures. CVE is a glossary that classifies vulnerabilities. The …"}
+{"idx": 6, "title": "CISA Presents Vision for the Common Vulnerabilities and Exposures ( CVE …", "date": "", "ddg_snippet": "Sep 10, 2025 · CISA believes the CVE program must be led with a commitment to conflict-free and vendor-neutral stewardship, broad multi-sector engagement, transparent processes, and …", "subpage_snippet": "", "source": "www.cisa.gov", "link": "https://www.cisa.gov/news-events/news/cisa-presents-vision-common-vulnerabilities-and-exposures-cve-program", "content": "Sep 10, 2025 · CISA believes the CVE program must be led with a commitment to conflict-free and vendor-neutral stewardship, broad multi-sector engagement, transparent processes, and …"}
+{"idx": 7, "title": "CVE Vault - CVE Database & Security Research Hub", "date": "", "ddg_snippet": "Comprehensive CVE database for cybersecurity research and vulnerability management. Search thousands of CVEs by severity, vendor, and keywords.", "subpage_snippet": "", "source": "cvevault.com", "link": "https://cvevault.com/", "content": "Comprehensive CVE database for cybersecurity research and vulnerability management. Search thousands of CVEs by severity, vendor, and keywords."}
+{"idx": 8, "title": "CVE Crowd | Crowd Intelligence on CVEs", "date": "", "ddg_snippet": "CVE-2025-21043として追跡されているこの重大なセキュリティ脆弱性は、Android 13以降を搭載したSamsungデバイスに影響し、MetaおよびWhatsAppのセキュリティチームによって8月13日に …", "subpage_snippet": "", "source": "api.cvecrowd.com", "link": "https://api.cvecrowd.com/", "content": "CVE-2025-21043として追跡されているこの重大なセキュリティ脆弱性は、Android 13以降を搭載したSamsungデバイスに影響し、MetaおよびWhatsAppのセキュリティチームによって8月13日に …"}
+{"idx": 9, "title": "NVD - Vulnerabilities", "date": "", "ddg_snippet": "The Common Vulnerabilities and Exposures (CVE) Program’s primary purpose is to uniquely identify vulnerabilities and to associate specific versions of code bases (e.g., software and shared …", "subpage_snippet": "", "source": "nvd.nist.gov", "link": "https://nvd.nist.gov/vuln", "content": "The Common Vulnerabilities and Exposures (CVE) Program’s primary purpose is to uniquely identify vulnerabilities and to associate specific versions of code bases (e.g., software and shared …"}
diff --git a/data/sampled_jsons/Chen_et_al.,_2023_reinforcement_learning_reference_trajectory_year_2023.jsonl b/data/sampled_jsons/Chen_et_al.,_2023_reinforcement_learning_reference_trajectory_year_2023.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..dbdecb261dca825bbc6631807ac51e25fbc876ed
--- /dev/null
+++ b/data/sampled_jsons/Chen_et_al.,_2023_reinforcement_learning_reference_trajectory_year_2023.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Reinforcement learning - Wikipedia", "date": "", "ddg_snippet": "The environment is typically stated in the form of a Markov decision process , as many reinforcement learning algorithms use dynamic programming ...", "subpage_snippet": "", "source": "en.wikipedia.org", "link": "https://en.wikipedia.org/wiki/Reinforcement_learning", "content": "The environment is typically stated in the form of a Markov decision process , as many reinforcement learning algorithms use dynamic programming ..."}
+{"idx": 1, "title": "(PDF) Reinforcement learning", "date": "", "ddg_snippet": "... reinforcement learning (RL) ... Large Language Models and RL Reinforcement Learning (RL) (Kaelbling, Littman, and Moore 1996; Sutton, Barto et al .", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/270960086_Reinforcement_learning", "content": "... reinforcement learning (RL) ... Large Language Models and RL Reinforcement Learning (RL) (Kaelbling, Littman, and Moore 1996; Sutton, Barto et al ."}
+{"idx": 2, "title": "Outcome-Based Online Reinforcement Learning: Algorithms and", "date": "", "ddg_snippet": "... reward structure also appears in the recent work on Reinforcement Learning from Human Feedback (RLHF) ( Chen et al ., 2022b , a ; Wu and Sun, 2023 ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2505.20268v2", "content": "... reward structure also appears in the recent work on Reinforcement Learning from Human Feedback (RLHF) ( Chen et al ., 2022b , a ; Wu and Sun, 2023 ..."}
+{"idx": 3, "title": "Any-point Trajectory Modeling for Policy Learning", "date": "", "ddg_snippet": "After training the model on an action-free video dataset, the predicted trajectories serve as effective guidance for learning visuomotor policies for ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2401.00025v3", "content": "After training the model on an action-free video dataset, the predicted trajectories serve as effective guidance for learning visuomotor policies for ..."}
+{"idx": 4, "title": "Reinforcement learning with model-based feedforward inputs for", "date": "", "ddg_snippet": "Second, reinforcement learning is often data hungry(Laskin et al ., 2020 ), although the required amount of training data depends on the task at hand.", "subpage_snippet": "", "source": "link.springer.com", "link": "https://link.springer.com/article/10.1007/s10514-023-10140-6", "content": "Second, reinforcement learning is often data hungry(Laskin et al ., 2020 ), although the required amount of training data depends on the task at hand."}
+{"idx": 5, "title": "Reinforcement learning — The Dan MacKinlay stable of", "date": "", "ddg_snippet": "Here’ s an intro to all of machine learning through a historical tale about one particular attempt to teach a machine (not a computer!) to play tic ...", "subpage_snippet": "", "source": "danmackinlay.name", "link": "https://danmackinlay.name/notebook/reinforcement_learning", "content": "Here’ s an intro to all of machine learning through a historical tale about one particular attempt to teach a machine (not a computer!) to play tic ..."}
+{"idx": 6, "title": "Rapid Vehicle Trajectory Prediction Based on Multi-Attention", "date": "", "ddg_snippet": "In the following sections, we provide a review of trajectory prediction algorithms based on deep learning methods [ 11 ].", "subpage_snippet": "", "source": "www.preprints.org", "link": "https://www.preprints.org/manuscript/202411.0644/v1", "content": "In the following sections, we provide a review of trajectory prediction algorithms based on deep learning methods [ 11 ]."}
+{"idx": 7, "title": "Newest 'notation' Questions - Artificial Intelligence", "date": "", "ddg_snippet": "In the context of Reinforcement Learning , I have seen that the policy $\\pi$ (for some algorithms) is nothing but a Neural Network architecture (for ...", "subpage_snippet": "", "source": "ai.stackexchange.com", "link": "https://ai.stackexchange.com/questions/tagged/notation", "content": "In the context of Reinforcement Learning , I have seen that the policy $\\pi$ (for some algorithms) is nothing but a Neural Network architecture (for ..."}
+{"idx": 8, "title": "Jiamou Liu | DeepAI", "date": "", "ddg_snippet": "Learnersourcing involves students generating and sharing learning resour... ... Few-shot learning (FSL) is an emergent paradigm of learning that ...", "subpage_snippet": "", "source": "deepai.org", "link": "https://deepai.org/profile/jiamou-liu", "content": "Learnersourcing involves students generating and sharing learning resour... ... Few-shot learning (FSL) is an emergent paradigm of learning that ..."}
+{"idx": 9, "title": "Most Influential IJCAI Papers (2024-05 Version) –", "date": "", "ddg_snippet": "MM-PCQA: Multi-Modal Learning for No- reference Point Cloud Quality Assessment IF:3 Related Papers Related Patents Related Grants Related Venues ...", "subpage_snippet": "", "source": "resources.paperdigest.org", "link": "https://resources.paperdigest.org/2024/05/most-influential-ijcai-papers-2024-05/", "content": "MM-PCQA: Multi-Modal Learning for No- reference Point Cloud Quality Assessment IF:3 Related Papers Related Patents Related Grants Related Venues ..."}
diff --git a/data/sampled_jsons/CoPINN_Cognitive_Physics-Informed_Neural_Networks_Figure_6_Section_3.5_beta_Helmholtz_equation_recom.jsonl b/data/sampled_jsons/CoPINN_Cognitive_Physics-Informed_Neural_Networks_Figure_6_Section_3.5_beta_Helmholtz_equation_recom.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..558fa165e8fe54638580ba4d6b48c49f1e4fd488
--- /dev/null
+++ b/data/sampled_jsons/CoPINN_Cognitive_Physics-Informed_Neural_Networks_Figure_6_Section_3.5_beta_Helmholtz_equation_recom.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "GitHub - siyuancncd/CoPINN: This is the official implementation of ...", "date": "", "ddg_snippet": "Physics - informed neural networks (PINNs) aim to constrain the outputs and gradients of deep learning models to satisfy specified governing physics equations , which have demonstrated significant potential for solving partial differential equations (PDEs). Although existing PINN methods have achieved pleasing performance, they always treat both easy and hard sample points indiscriminately ...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/siyuancncd/CoPINN", "content": "Physics - informed neural networks (PINNs) aim to constrain the outputs and gradients of deep learning models to satisfy specified governing physics equations , which have demonstrated significant potential for solving partial differential equations (PDEs). Although existing PINN methods have achieved pleasing performance, they always treat both easy and hard sample points indiscriminately ..."}
+{"idx": 1, "title": "CoPINN: Cognitive Physics-Informed Neural Networks", "date": "", "ddg_snippet": "Abstract Physics - informed neural networks (PINN) aim to constrain the outputs and gradients of deep learn-ing models to satisfy specified governing physics equations , which have demonstrated significant potential for solving partial differential equations (PDEs). Although existing PINN methods have achieved pleasing performance, they always treat both easy and hard sample points ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=4vAa0A98xI", "content": "Abstract Physics - informed neural networks (PINN) aim to constrain the outputs and gradients of deep learn-ing models to satisfy specified governing physics equations , which have demonstrated significant potential for solving partial differential equations (PDEs). Although existing PINN methods have achieved pleasing performance, they always treat both easy and hard sample points ..."}
+{"idx": 2, "title": "CoPINN: Cognitive Physics-Informed Neural Networks", "date": "", "ddg_snippet": "The paper introduces a new method called Cognitive Physics - Informed Neural Networks ( CoPINN ) to improve how computers solve complex math problems known as partial differential equations (PDEs). Un...", "subpage_snippet": "", "source": "bytez.com", "link": "https://bytez.com/docs/icml/46458/paper", "content": "The paper introduces a new method called Cognitive Physics - Informed Neural Networks ( CoPINN ) to improve how computers solve complex math problems known as partial differential equations (PDEs). Un..."}
+{"idx": 3, "title": "PDF CoPINN/CoPINN.pdf at main · siyuancncd/CoPINN · GitHub", "date": "", "ddg_snippet": "This is the official implementation of \" CoPINN : Cognitive Physics - informed Neural Network \" (ICML 2025, Spotlight) - CoPINN / CoPINN .pdf at main · siyuancncd/ CoPINN", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/siyuancncd/CoPINN/blob/main/CoPINN.pdf", "content": "This is the official implementation of \" CoPINN : Cognitive Physics - informed Neural Network \" (ICML 2025, Spotlight) - CoPINN / CoPINN .pdf at main · siyuancncd/ CoPINN"}
+{"idx": 4, "title": "Machine-Learning/Physics Informed Neural Networks with Python ... - GitHub", "date": "", "ddg_snippet": "Slide 1: Introduction to Physics Informed Neural Networks (PINNs) Physics Informed Neural Networks (PINNs) are a novel approach that combines the power of neural networks with the fundamental laws of physics . They aim to solve complex physical problems by incorporating physical constraints into the learning process, resulting in more accurate and physically consistent predictions.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/xbeat/Machine-Learning/blob/main/Physics+Informed+Neural+Networks+with+Python.md", "content": "Slide 1: Introduction to Physics Informed Neural Networks (PINNs) Physics Informed Neural Networks (PINNs) are a novel approach that combines the power of neural networks with the fundamental laws of physics . They aim to solve complex physical problems by incorporating physical constraints into the learning process, resulting in more accurate and physically consistent predictions."}
+{"idx": 5, "title": "[2405.08111] Conformalized Physics-Informed Neural Networks", "date": "", "ddg_snippet": "Physics - informed neural networks (PINNs) are an influential method of solving differential equations and estimating their parameters given data. However, since they make use of neural networks , they provide only a point estimate of differential equation parameters, as well as the solution at any given point, without any measure of uncertainty. Ensemble and Bayesian methods have been previously ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2405.08111", "content": "Physics - informed neural networks (PINNs) are an influential method of solving differential equations and estimating their parameters given data. However, since they make use of neural networks , they provide only a point estimate of differential equation parameters, as well as the solution at any given point, without any measure of uncertainty. Ensemble and Bayesian methods have been previously ..."}
+{"idx": 6, "title": "CoPINN: Cognitive Physics-Informed Neural Networks", "date": "", "ddg_snippet": "Physics - informed neural networks (PINNs) aim to constrain the outputs and gradients of deep learning models to satisfy specified governing physics equations , which have demonstrated significant potential for solving partial differential equations (PDEs). Although existing PINN methods have achieved pleasing performance, they always treat both easy and hard sample points indiscriminately ...", "subpage_snippet": "", "source": "chatpaper.com", "link": "https://chatpaper.com/chatpaper/paper/165180?from=subpath-search", "content": "Physics - informed neural networks (PINNs) aim to constrain the outputs and gradients of deep learning models to satisfy specified governing physics equations , which have demonstrated significant potential for solving partial differential equations (PDEs). Although existing PINN methods have achieved pleasing performance, they always treat both easy and hard sample points indiscriminately ..."}
+{"idx": 7, "title": "CoPINN/README.md at main · siyuancncd/CoPINN · GitHub", "date": "", "ddg_snippet": "About This is the official implementation of \" CoPINN : Cognitive Physics - informed Neural Network \" (ICML 2025, PyTorch Code) ‼️ I'm actively seeking a PhD position for Fall 2026 entry. If you believe my background aligns with your research needs, please feel free to contact me via email at ...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/siyuancncd/CoPINN/blob/main/README.md", "content": "About This is the official implementation of \" CoPINN : Cognitive Physics - informed Neural Network \" (ICML 2025, PyTorch Code) ‼️ I'm actively seeking a PhD position for Fall 2026 entry. If you believe my background aligns with your research needs, please feel free to contact me via email at ..."}
+{"idx": 8, "title": "CoPINN: Cognitive Physics-Informed Neural Networks", "date": "", "ddg_snippet": "Physics - informed neural networks (PINNs) aim to constrain the outputs and gradients of deep learning models to satisfy specified governing physics equations , which have demonstrated significant...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=4vAa0A98xI", "content": "Physics - informed neural networks (PINNs) aim to constrain the outputs and gradients of deep learning models to satisfy specified governing physics equations , which have demonstrated significant..."}
+{"idx": 9, "title": "[2204.11144] Competitive Physics Informed Networks - arXiv.org", "date": "", "ddg_snippet": "This strategy is called \" physics - informed neural networks \" (PINNs), but it currently cannot produce high-accuracy solutions, typically attaining about 0.1% relative error.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2204.11144", "content": "This strategy is called \" physics - informed neural networks \" (PINNs), but it currently cannot produce high-accuracy solutions, typically attaining about 0.1% relative error."}
diff --git a/data/sampled_jsons/CoPINN_Cognitive_Physics-Informed_Neural_Networks_Section_2.4_sample_difficulty_evaluation_mechanism.jsonl b/data/sampled_jsons/CoPINN_Cognitive_Physics-Informed_Neural_Networks_Section_2.4_sample_difficulty_evaluation_mechanism.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..dd9053fb6b878debfa22efef964ef0331f5f0820
--- /dev/null
+++ b/data/sampled_jsons/CoPINN_Cognitive_Physics-Informed_Neural_Networks_Section_2.4_sample_difficulty_evaluation_mechanism.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "bitzhangcy/Neural-PDE-Solver", "date": "", "ddg_snippet": "This is an open-source repository for Neural -PDE-Solver, a curated collection of literature on solving Partial Differential Equations (PDEs) using Neural ...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/bitzhangcy/Neural-PDE-Solver", "content": "This is an open-source repository for Neural -PDE-Solver, a curated collection of literature on solving Partial Differential Equations (PDEs) using Neural ..."}
+{"idx": 1, "title": "Knowledge Modelling and Learning through Cognitive ...", "date": "", "ddg_snippet": "following sections , the sample , procedure, materials, and network analysis are described. ... belong to the neural networks method and ensemble methods.", "subpage_snippet": "", "source": "mdpi-res.com", "link": "https://mdpi-res.com/bookfiles/book/5650/Knowledge_Modelling_and_Learning_through_Cognitive_Networks.pdf?v=1746147819", "content": "following sections , the sample , procedure, materials, and network analysis are described. ... belong to the neural networks method and ensemble methods."}
+{"idx": 2, "title": "Cognitive Development (Part I) - New Perspectives on ...", "date": "", "ddg_snippet": "11 May 2017 — This chapter provides an overview of research using constructive neural networks to simulate phenomena in cognitive development.", "subpage_snippet": "", "source": "www.cambridge.org", "link": "https://www.cambridge.org/core/books/new-perspectives-on-human-development/cognitive-development/8C91FE0D753BDB98264F76F4864CE044", "content": "11 May 2017 — This chapter provides an overview of research using constructive neural networks to simulate phenomena in cognitive development."}
+{"idx": 3, "title": "Neural Information Processing", "date": "", "ddg_snippet": "3 Mar 2025 — Welcome to the 31st International Conference on Neural Information Processing. (ICONIP 2024) of the Asia-Pacific Neural Network Society (APNNS), ...", "subpage_snippet": "", "source": "link.springer.com", "link": "https://link.springer.com/content/pdf/10.1007/978-981-96-6579-2.pdf", "content": "3 Mar 2025 — Welcome to the 31st International Conference on Neural Information Processing. (ICONIP 2024) of the Asia-Pacific Neural Network Society (APNNS), ..."}
+{"idx": 4, "title": "NeurIPS 2024 Friday 12/13", "date": "", "ddg_snippet": "When applied to Physics - Informed Neural Networks ( PINNs ), our method provides >1000 × speed-up and >30 × memory reduction over randomization with first ...", "subpage_snippet": "", "source": "neurips.cc", "link": "https://neurips.cc/virtual/2024/day/12/13", "content": "When applied to Physics - Informed Neural Networks ( PINNs ), our method provides >1000 × speed-up and >30 × memory reduction over randomization with first ..."}
+{"idx": 5, "title": "Synergies between machine learning and reasoning", "date": "", "ddg_snippet": "by I Baaj · 2024 · Cited by 9 — Thus, for instance, Physics - Informed Neural Networks ( PINNs ) attempt both to optimize the fit to the training data (e.g. as measured by the ...", "subpage_snippet": "", "source": "www.sciencedirect.com", "link": "https://www.sciencedirect.com/science/article/am/pii/S0888613X24000938", "content": "by I Baaj · 2024 · Cited by 9 — Thus, for instance, Physics - Informed Neural Networks ( PINNs ) attempt both to optimize the fit to the training data (e.g. as measured by the ..."}
+{"idx": 6, "title": "A review on model-based design of experiments for ...", "date": "", "ddg_snippet": "by M Geremia · 2026 — Appropriate physics -based assumptions can provide a deep understanding of the underlying phenomena within the system under investigation and enable predicting ...", "subpage_snippet": "", "source": "www.sciencedirect.com", "link": "https://www.sciencedirect.com/science/article/pii/S0009250925011686", "content": "by M Geremia · 2026 — Appropriate physics -based assumptions can provide a deep understanding of the underlying phenomena within the system under investigation and enable predicting ..."}
+{"idx": 7, "title": "Deep Learning: An Introduction for Applied Mathematicians", "date": "", "ddg_snippet": "by CF Higham · 2019 · Cited by 384 — This article provides a very brief introduction to the basic ideas that underlie deep learning from an applied mathematics perspective.", "subpage_snippet": "", "source": "epubs.siam.org", "link": "https://epubs.siam.org/doi/full/10.1137/18M1165748", "content": "by CF Higham · 2019 · Cited by 384 — This article provides a very brief introduction to the basic ideas that underlie deep learning from an applied mathematics perspective."}
+{"idx": 8, "title": "What Is Deep Learning, and How Should We Evaluate Its ...", "date": "", "ddg_snippet": "14 Dec 2023 — This chapter explains the basic methodology of deep learning research and reviews its major achievements and criticisms.", "subpage_snippet": "", "source": "academic.oup.com", "link": "https://academic.oup.com/book/55239/chapter/428593948", "content": "14 Dec 2023 — This chapter explains the basic methodology of deep learning research and reviews its major achievements and criticisms."}
+{"idx": 9, "title": "NIST.IR.8367.pdf", "date": "", "ddg_snippet": "by DA Broniatowski · 2021 · Cited by 111 — Interpretable machine learning is concerned with helping humans generate interpretations of data and model output. Thus, we review literature in human ...", "subpage_snippet": "", "source": "nvlpubs.nist.gov", "link": "https://nvlpubs.nist.gov/nistpubs/ir/2021/NIST.IR.8367.pdf", "content": "by DA Broniatowski · 2021 · Cited by 111 — Interpretable machine learning is concerned with helping humans generate interpretations of data and model output. Thus, we review literature in human ..."}
diff --git a/data/sampled_jsons/Contrastive_CRL_real_data_noise_assumptions_violated.jsonl b/data/sampled_jsons/Contrastive_CRL_real_data_noise_assumptions_violated.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..eb1c0a7c6e8e3d9ab1f6ead40e1b7283caf3801c
--- /dev/null
+++ b/data/sampled_jsons/Contrastive_CRL_real_data_noise_assumptions_violated.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Contrastive Representations for Temporal Reasoning", "date": "", "ddg_snippet": "18 Aug 2025 — Contrastive learning acquires representations by pulling representations of similar data points, i.e., ones that belong to the same underlying ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2508.13113v1", "content": "18 Aug 2025 — Contrastive learning acquires representations by pulling representations of similar data points, i.e., ones that belong to the same underlying ..."}
+{"idx": 1, "title": "Contrastive Representations for Temporal Reasoning", "date": "", "ddg_snippet": "by A Ziarko · 2025 — Contrastive learning acquires representations by pulling representations of similar data points, i.e., ones that belong to the same underlying ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2508.13113?", "content": "by A Ziarko · 2025 — Contrastive learning acquires representations by pulling representations of similar data points, i.e., ones that belong to the same underlying ..."}
+{"idx": 2, "title": "Embedding Dimension of Contrastive Learning and", "date": "", "ddg_snippet": "The authors should reflect on how these assumptions might be violated in practice and what the implications would be. • The authors should reflect on the ...", "subpage_snippet": "", "source": "neurips.cc", "link": "https://neurips.cc/virtual/2024/poster/95861", "content": "The authors should reflect on how these assumptions might be violated in practice and what the implications would be. • The authors should reflect on the ..."}
+{"idx": 3, "title": "Membership Inference Attacks Against Synthetic Health Data", "date": "", "ddg_snippet": "by Z Zhang · 2021 · Cited by 97 — We introduce a framework for effective membership inference against synthetic health data without specific assumptions about the generative model or a well- ...", "subpage_snippet": "", "source": "pmc.ncbi.nlm.nih.gov", "link": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8766950/", "content": "by Z Zhang · 2021 · Cited by 97 — We introduce a framework for effective membership inference against synthetic health data without specific assumptions about the generative model or a well- ..."}
+{"idx": 4, "title": "Causal Component Analysis", "date": "", "ddg_snippet": "by W Liang · Cited by 63 — In the context of state-of-the-art in CRL , there is a complete lack of similar identifiability results without stronger assumptions on the mixing function, or ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=HszLRiHyfO", "content": "by W Liang · Cited by 63 — In the context of state-of-the-art in CRL , there is a complete lack of similar identifiability results without stronger assumptions on the mixing function, or ..."}
+{"idx": 5, "title": "Score-based Causal Representation Learning: Linear and ...", "date": "", "ddg_snippet": "by B Varici · 2025 · Cited by 4 — we show that this assumption is violated for linear additive noise models as follows. First, we quote their assumption : Linear interventional faithfulness. 90 pages", "subpage_snippet": "", "source": "www.jmlr.org", "link": "https://www.jmlr.org/papers/volume26/24-0194/24-0194.pdf", "content": "by B Varici · 2025 · Cited by 4 — we show that this assumption is violated for linear additive noise models as follows. First, we quote their assumption : Linear interventional faithfulness. 90 pages"}
+{"idx": 6, "title": "ICML 2025 Orals", "date": "", "ddg_snippet": "Constrained Reinforcement Learning ( CRL ) aims to maximize cumulative rewards while satisfying constraints. However, existing CRL algorithms often encounter ...", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/events/oral", "content": "Constrained Reinforcement Learning ( CRL ) aims to maximize cumulative rewards while satisfying constraints. However, existing CRL algorithms often encounter ..."}
+{"idx": 7, "title": "Multi-Domain Causal Representation Learning via Weak ...", "date": "", "ddg_snippet": "by K Ahuja · 2024 · Cited by 19 — This assumption is often violated : e.g. the causal relationships between the latents can have different causal directions in two images, where a cat chases a ...", "subpage_snippet": "", "source": "proceedings.mlr.press", "link": "https://proceedings.mlr.press/v238/ahuja24a/ahuja24a.pdf", "content": "by K Ahuja · 2024 · Cited by 19 — This assumption is often violated : e.g. the causal relationships between the latents can have different causal directions in two images, where a cat chases a ..."}
+{"idx": 8, "title": "Linear Causal Representation Learning from Unknown ...", "date": "", "ddg_snippet": "by B Varıcı · Cited by 4 — The authors should reflect on how these assumptions might be violated in practice and what the implications would be. • The authors should reflect on the ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=weemASPtzg", "content": "by B Varıcı · Cited by 4 — The authors should reflect on how these assumptions might be violated in practice and what the implications would be. • The authors should reflect on the ..."}
+{"idx": 9, "title": "Causal Representation Learning from General Environments ...", "date": "", "ddg_snippet": "by I Ng — Causal representation learning aims to re- cover the latent causal variables and their causal relations, typically represented by di-.", "subpage_snippet": "", "source": "raw.githubusercontent.com", "link": "https://raw.githubusercontent.com/mlresearch/v258/main/assets/ng25a/ng25a.pdf", "content": "by I Ng — Causal representation learning aims to re- cover the latent causal variables and their causal relations, typically represented by di-."}
diff --git a/data/sampled_jsons/Corollary_2_lower_bound_expected_samples_E[T_i]_epistemic_limits_passive_data_collection_Nickel.jsonl b/data/sampled_jsons/Corollary_2_lower_bound_expected_samples_E[T_i]_epistemic_limits_passive_data_collection_Nickel.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..4266402736f8bdb60a898c4f8573089e8993d636
--- /dev/null
+++ b/data/sampled_jsons/Corollary_2_lower_bound_expected_samples_E[T_i]_epistemic_limits_passive_data_collection_Nickel.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Corollary - Wikipedia", "date": "", "ddg_snippet": "In mathematics and logic, a corollary is a theorem of less importance which can be readily deduced from a previous, more notable statement. A corollary could, for instance, be a proposition which is incidentally proved while proving another propositi...", "subpage_snippet": "", "source": "en.wikipedia.org", "link": "https://en.wikipedia.org/wiki/Corollary", "content": "In mathematics and logic, a corollary is a theorem of less importance which can be readily deduced from a previous, more notable statement. A corollary could, for instance, be a proposition which is incidentally proved while proving another propositi..."}
+{"idx": 1, "title": "(PDF) On the Benefits of Active Data Collection in Operator Learning", "date": "", "ddg_snippet": "under any passive data collection strategy. That is, the lower bound does not vanish even as. n→ ∞. Collectively, Theorems 1and 2 establish a clear advantage of active data collection .", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/385291976_On_the_Benefits_of_Active_Data_Collection_in_Operator_Learning", "content": "under any passive data collection strategy. That is, the lower bound does not vanish even as. n→ ∞. Collectively, Theorems 1and 2 establish a clear advantage of active data collection ."}
+{"idx": 2, "title": "std:: lower _ bound - cppreference.com", "date": "", "ddg_snippet": "Metaprogramming library. General utilities library. Containers library. Iterators library. Ranges library. Algorithms library. Strings library. Text processing library. Numerics library. Date and time library. Input/output library. Filesystem library...", "subpage_snippet": "", "source": "en.cppreference.com", "link": "https://en.cppreference.com/w/cpp/algorithm/lower_bound.html", "content": "Metaprogramming library. General utilities library. Containers library. Iterators library. Ranges library. Algorithms library. Strings library. Text processing library. Numerics library. Date and time library. Input/output library. Filesystem library..."}
+{"idx": 3, "title": "No free delivery service", "date": "", "ddg_snippet": "No free delivery service Epistemic limits of passive data collection .", "subpage_snippet": "", "source": "proceedings.neurips.cc", "link": "https://proceedings.neurips.cc/paper_files/paper/2024/file/b97fc02c9e536d68300d82be05c23aa2-Paper-Conference.pdf", "content": "No free delivery service Epistemic limits of passive data collection ."}
+{"idx": 4, "title": "A Tight Quantum Algorithm for Multiple Collision Search", "date": "", "ddg_snippet": "When one searches for a single collision, the known quantum algorithms match the query lower bound . This is not the case for the problem of finding multiple collisions, despite its regular appearance as a sub-component in sieving-type algorithms.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2509.13909v1", "content": "When one searches for a single collision, the known quantum algorithms match the query lower bound . This is not the case for the problem of finding multiple collisions, despite its regular appearance as a sub-component in sieving-type algorithms."}
+{"idx": 5, "title": "Advanced Weakest Precondition Calculi for Probabilistic Programs", "date": "", "ddg_snippet": "We show that lower bounds on expected values are not computable but computably enumerable, whereas upper bounds are not computably enumerable. Corollary 4.23 (Angelic Bound Demonic Preexpectations): Let C ∈ pGCL, f ∈ E , and g ∈ E ≤1. Then.", "subpage_snippet": "", "source": "publications.rwth-aachen.de", "link": "https://publications.rwth-aachen.de/record/755408/files/755408.pdf", "content": "We show that lower bounds on expected values are not computable but computably enumerable, whereas upper bounds are not computably enumerable. Corollary 4.23 (Angelic Bound Demonic Preexpectations): Let C ∈ pGCL, f ∈ E , and g ∈ E ≤1. Then."}
+{"idx": 6, "title": "Huawei Watch Ultimate 2 Review: 150m Diving & Underwater Text", "date": "", "ddg_snippet": "Can the Watch Ultimate 2 replace a dive computer? For recreational and technical divers, it provides many dive computer functions, including depth tracking, no-decompression limits , and safety stop timers.", "subpage_snippet": "", "source": "www.superfashion.us", "link": "https://www.superfashion.us/huawei-watch-ultimate-2-review/", "content": "Can the Watch Ultimate 2 replace a dive computer? For recreational and technical divers, it provides many dive computer functions, including depth tracking, no-decompression limits , and safety stop timers."}
+{"idx": 7, "title": "Двигатель 8NR-FTS 1. 2 - характеристики, масло, ресурс, фильтры", "date": "", "ddg_snippet": "Описание двигателя Toyota 8NR-FTS 1. 2 (двигатель C-HR, Corolla , Auris и др.). Его устройство и характеристики, ресурс, отзывы, общие проблемы, какое масло лить, артикулы расходников.", "subpage_snippet": "", "source": "www.motorhunter.ru", "link": "https://www.motorhunter.ru/engine/toyota/8nr-fts/", "content": "Описание двигателя Toyota 8NR-FTS 1. 2 (двигатель C-HR, Corolla , Auris и др.). Его устройство и характеристики, ресурс, отзывы, общие проблемы, какое масло лить, артикулы расходников."}
+{"idx": 8, "title": "Toyota Corolla ( E 120) — верная \"коровка\". — DRIVE 2", "date": "", "ddg_snippet": "Тойота Королла считается бестселлером автомобилестроения и занесена в Книгу Рекордов Гиннеса, как самая продаваемая модель в мире. Toyota Corolla девятого поколения, получившая обозначение Е 120, уви…", "subpage_snippet": "", "source": "www.drive2.ru", "link": "https://www.drive2.ru/b/4899916394579169027/", "content": "Тойота Королла считается бестселлером автомобилестроения и занесена в Книгу Рекордов Гиннеса, как самая продаваемая модель в мире. Toyota Corolla девятого поколения, получившая обозначение Е 120, уви…"}
+{"idx": 9, "title": "autozone.com/batteries-starting-and-charging/battery", "date": "", "ddg_snippet": "...BCI Group 47 690 CCA H5-AGM-600 for Chevrolet Cruze Limited .", "subpage_snippet": "", "source": "www.autozone.com", "link": "https://www.autozone.com/batteries-starting-and-charging/battery", "content": "...BCI Group 47 690 CCA H5-AGM-600 for Chevrolet Cruze Limited ."}
diff --git a/data/sampled_jsons/DIKE_classification_accuracy_vs_GPT-4_zero-shot_Figure_4a_percentage_points_Checks-and-Balances_Fram_year_2024.jsonl b/data/sampled_jsons/DIKE_classification_accuracy_vs_GPT-4_zero-shot_Figure_4a_percentage_points_Checks-and-Balances_Fram_year_2024.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..3b2769c7d44b5c0041b375418f9eec2f8d36ebf0
--- /dev/null
+++ b/data/sampled_jsons/DIKE_classification_accuracy_vs_GPT-4_zero-shot_Figure_4a_percentage_points_Checks-and-Balances_Fram_year_2024.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "A Checks - and - Balances Framework for Context - Aware Ethical AI ...", "date": "", "ddg_snippet": "Figure 4 a shows that Dike ’s classification accuracy sur-passes GPT - 4 ’s zero - shot method by 11.3 percentage points , confirming the effectiveness of emotion-mediated behavior classification . The 5% error bar reflects the inherent...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2502.00136", "content": "Figure 4 a shows that Dike ’s classification accuracy sur-passes GPT - 4 ’s zero - shot method by 11.3 percentage points , confirming the effectiveness of emotion-mediated behavior classification . The 5% error bar reflects the inherent..."}
+{"idx": 1, "title": "A Three-Branch Checks - and - Balances Framework", "date": "", "ddg_snippet": "Figure 2a demonstrates that DIKE ’s classification accuracy surpasses GPT - 4 ’s zero - shot method by 11.3 percentage points , confirming the effectiveness of DIKE ’s detailed emotion-behavior mapping.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=o2afWIxjKD", "content": "Figure 2a demonstrates that DIKE ’s classification accuracy surpasses GPT - 4 ’s zero - shot method by 11.3 percentage points , confirming the effectiveness of DIKE ’s detailed emotion-behavior mapping."}
+{"idx": 2, "title": "(PDF) Checks - and - Balances Framework for Context - Aware Ethical ...", "date": "", "ddg_snippet": "A novel checks - and - balances architecture for ethical . alignment that maintains separation between knowledge. Figure 4 a shows that. Dike . ’s classification accuracy sur-. passes GPT - 4 ’s zero - shot method by 11.3 percentage points", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/380515639_Checks-and-Balances_Framework_for_Context-Aware_Ethical_AI_Alignment", "content": "A novel checks - and - balances architecture for ethical . alignment that maintains separation between knowledge. Figure 4 a shows that. Dike . ’s classification accuracy sur-. passes GPT - 4 ’s zero - shot method by 11.3 percentage points"}
+{"idx": 3, "title": "Сравнение: Qwen 3 Max Preview против GPT -5. Тестирование...", "date": "", "ddg_snippet": "GPT-5 хорошо справляется с этой задачей. Хотя сравнение не вполне справедливо, т.к. GPT-5 может использовать режим рассуждений а Qwen 3 Max пока что нет. Среди нерассуждающих моделей результат Qwen 3 Max неплох, так как только GPT - 4 .5 в свое время умела...", "subpage_snippet": "", "source": "meanotek.io", "link": "https://meanotek.io/blog-entryqwen3-max-preview-vs-gpt5", "content": "GPT-5 хорошо справляется с этой задачей. Хотя сравнение не вполне справедливо, т.к. GPT-5 может использовать режим рассуждений а Qwen 3 Max пока что нет. Среди нерассуждающих моделей результат Qwen 3 Max неплох, так как только GPT - 4 .5 в свое время умела..."}
+{"idx": 4, "title": "ICML Poster A Three-Branch Checks - and - Balances Framework for ...", "date": "", "ddg_snippet": "This paper introduces a three-branch checks - and - balances framework for ethical alignment of Large Language Models (LLMs), inspired by the idea of collaborative intelligence.", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/poster/46461", "content": "This paper introduces a three-branch checks - and - balances framework for ethical alignment of Large Language Models (LLMs), inspired by the idea of collaborative intelligence."}
+{"idx": 5, "title": "Inclusive prompt engineering for large language models: a modular...", "date": "", "ddg_snippet": "Techniques evolved from zero - shot and few-shot prompting to more structured methods like Chain-of-Thought (CoT) and memory augmentation. Despite their effectiveness, these methods are often evaluated in isolation.", "subpage_snippet": "", "source": "link.springer.com", "link": "https://link.springer.com/article/10.1007/s10462-025-11330-7", "content": "Techniques evolved from zero - shot and few-shot prompting to more structured methods like Chain-of-Thought (CoT) and memory augmentation. Despite their effectiveness, these methods are often evaluated in isolation."}
+{"idx": 6, "title": "JMIR Medical Informatics - Leveraging GPT - 4 o for Automated...", "date": "", "ddg_snippet": "Leveraging GPT - 4 o for Automated Extraction and Categorization of CAD-RADS Features From Free-Text Coronary CT Angiography Reports: Diagnostic Study.", "subpage_snippet": "", "source": "medinform.jmir.org", "link": "https://medinform.jmir.org/2025/1/e70967", "content": "Leveraging GPT - 4 o for Automated Extraction and Categorization of CAD-RADS Features From Free-Text Coronary CT Angiography Reports: Diagnostic Study."}
+{"idx": 7, "title": "An interactive demo for developers to try the new text-to-speech model...", "date": "", "ddg_snippet": "Delivery: Fast-paced and dynamic, with rising intonation to build momentum and keep engagement high.", "subpage_snippet": "", "source": "www.openai.fm", "link": "https://www.openai.fm/", "content": "Delivery: Fast-paced and dynamic, with rising intonation to build momentum and keep engagement high."}
+{"idx": 8, "title": "(PDF) A Three-Branch Checks - and - Balances Framework for ...", "date": "", "ddg_snippet": "AI Safety is an important issue, but the current solution with RLHF suffers from several shortcomings including subjective feedback, feedback hacking, and the forgetting effect. This work, which will be presented at NeurIPS this week, proposes a paradigm shift: using three LLM modules to...", "subpage_snippet": "", "source": "www.linkedin.com", "link": "https://www.linkedin.com/posts/edward-y-chang-218b182_pdf-a-three-branch-checks-and-balances-activity-7272792943923458050-RO6k", "content": "AI Safety is an important issue, but the current solution with RLHF suffers from several shortcomings including subjective feedback, feedback hacking, and the forgetting effect. This work, which will be presented at NeurIPS this week, proposes a paradigm shift: using three LLM modules to..."}
+{"idx": 9, "title": "GPT - 4 : что умеет нейросеть и как ею пользоваться", "date": "", "ddg_snippet": "Что нужно знать о GPT - 4 . Как пользоваться GPT - 4 . Чем GPT - 4 лучше ChatGPT. Какие возможности есть у нейросети. Где попробовать GPT - 4 .", "subpage_snippet": "", "source": "t-j.ru", "link": "https://t-j.ru/gpt4/", "content": "Что нужно знать о GPT - 4 . Как пользоваться GPT - 4 . Чем GPT - 4 лучше ChatGPT. Какие возможности есть у нейросети. Где попробовать GPT - 4 ."}
diff --git a/data/sampled_jsons/Davies_et_al._2021_Nature_knot_theory_representation_theory_mathematical_focus.jsonl b/data/sampled_jsons/Davies_et_al._2021_Nature_knot_theory_representation_theory_mathematical_focus.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..d01ff594337ebccba8343f5bb50a879c5e7d6ed7
--- /dev/null
+++ b/data/sampled_jsons/Davies_et_al._2021_Nature_knot_theory_representation_theory_mathematical_focus.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Deep Learning and Mathematical Intuition", "date": "", "ddg_snippet": "by E Davis · 2021 · Cited by 4 — (The account of the math in ( Davies et al . 2021 ) is likewise noticeably sketchier for the representation theory than the knot theory , for the ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2112.04324", "content": "by E Davis · 2021 · Cited by 4 — (The account of the math in ( Davies et al . 2021 ) is likewise noticeably sketchier for the representation theory than the knot theory , for the ..."}
+{"idx": 1, "title": "Comment on (Romera-Paredes et al., 2023), “Mathematical ...", "date": "", "ddg_snippet": "18 Dec 2023 — Deep learning technology is used to suggest conjectures in representation theory and knot theory which led to the proofs of theorems. ( Davies et ...", "subpage_snippet": "", "source": "cs.nyu.edu", "link": "https://cs.nyu.edu/~davise/papers/FunSearchComment.pdf", "content": "18 Dec 2023 — Deep learning technology is used to suggest conjectures in representation theory and knot theory which led to the proofs of theorems. ( Davies et ..."}
+{"idx": 2, "title": "Machine learning of knot topology in non-Hermitian band ...", "date": "", "ddg_snippet": "by J Chen · 2024 · Cited by 12 — Our study shows significant potential of machine learning in classification of knots , braid groups, and non-Hermitian topological phases.", "subpage_snippet": "", "source": "www.nature.com", "link": "https://www.nature.com/articles/s42005-024-01710-w", "content": "by J Chen · 2024 · Cited by 12 — Our study shows significant potential of machine learning in classification of knots , braid groups, and non-Hermitian topological phases."}
+{"idx": 3, "title": "Exploring Representations and Inductive Bias for Machine ...", "date": "", "ddg_snippet": "by N Driggs · 2025 — Knot theory is a branch of mathematics that studies embeddings of the circle in R3 that are equivalent up to ambient isotopy. A link is a knot ...", "subpage_snippet": "", "source": "scholarsarchive.byu.edu", "link": "https://scholarsarchive.byu.edu/cgi/viewcontent.cgi?article=11822&context=etd", "content": "by N Driggs · 2025 — Knot theory is a branch of mathematics that studies embeddings of the circle in R3 that are equivalent up to ambient isotopy. A link is a knot ..."}
+{"idx": 4, "title": "Advancing mathematics by guiding human intuition with AI", "date": "", "ddg_snippet": "by A Davies · 2021 · Cited by 686 — Interactive notebooks to regenerate the results for both knot theory and representation theory have been made available for download at https:// ...", "subpage_snippet": "", "source": "www.nature.com", "link": "https://www.nature.com/articles/s41586-021-04086-x", "content": "by A Davies · 2021 · Cited by 686 — Interactive notebooks to regenerate the results for both knot theory and representation theory have been made available for download at https:// ..."}
+{"idx": 5, "title": "Finding Increasingly Large Extremal Graphs with ...", "date": "", "ddg_snippet": "by A Mehrabian · 2023 · Cited by 13 — search problems across the fields of representation theory, knot theory, graph theory, and matrix algebra [Davies et al., 2021;. Fawzi et al., 2022; Wagner, ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2311.03583", "content": "by A Mehrabian · 2023 · Cited by 13 — search problems across the fields of representation theory, knot theory, graph theory, and matrix algebra [Davies et al., 2021;. Fawzi et al., 2022; Wagner, ..."}
+{"idx": 6, "title": "On scientific understanding with artificial intelligence", "date": "", "ddg_snippet": "by M Krenn · 2022 · Cited by 359 — ... knot theory , which allowed mathematicians to conjecture and prove new theorems. ... et al . Simulation intelligence: Towards a new generation of ...", "subpage_snippet": "", "source": "www.nature.com", "link": "https://www.nature.com/articles/s42254-022-00518-3", "content": "by M Krenn · 2022 · Cited by 359 — ... knot theory , which allowed mathematicians to conjecture and prove new theorems. ... et al . Simulation intelligence: Towards a new generation of ..."}
+{"idx": 7, "title": "AI, Robot Neuroscientist: Reimagining Hypothesis ...", "date": "", "ddg_snippet": "by J Shang · Cited by 1 — Davies et al. (2021) used this approach to help mathematicians formulate two conjectures , one in knot theory and the other in representation theory, which ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=CllNd4XWVF", "content": "by J Shang · Cited by 1 — Davies et al. (2021) used this approach to help mathematicians formulate two conjectures , one in knot theory and the other in representation theory, which ..."}
+{"idx": 8, "title": "Working with machines in mathematics", "date": "", "ddg_snippet": "15 May 2024 — ... natural slope, in the field of knot theory . It demonstrates how even a relatively simple connection in a well studied area can go unnoticed ...", "subpage_snippet": "", "source": "www.ams.org", "link": "https://www.ams.org/journals/bull/2024-61-03/S0273-0979-2024-01843-6/viewer", "content": "15 May 2024 — ... natural slope, in the field of knot theory . It demonstrates how even a relatively simple connection in a well studied area can go unnoticed ..."}
+{"idx": 9, "title": "NOT CONSTRUCTING RAMSEY GRAPHS USING DEEP ...", "date": "", "ddg_snippet": "by D Berghaus — We consider the problem of constructing Ramsey graphs using deep reinforcement learning. We introduce a novel permutation invariant architecture that ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=ODY9UitugC", "content": "by D Berghaus — We consider the problem of constructing Ramsey graphs using deep reinforcement learning. We introduce a novel permutation invariant architecture that ..."}
diff --git a/data/sampled_jsons/Dike_Eris_components_checks_balances_framework_ethical_AI_alignment_year_2024.jsonl b/data/sampled_jsons/Dike_Eris_components_checks_balances_framework_ethical_AI_alignment_year_2024.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..c956ddb9446d66d6a3d60a046ce7b891e100f87d
--- /dev/null
+++ b/data/sampled_jsons/Dike_Eris_components_checks_balances_framework_ethical_AI_alignment_year_2024.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "(PDF) Checks -and- Balances Framework for Context-Aware Ethical ...", "date": "", "ddg_snippet": "This paper introduces a checks -and- balances framework for ethical alignment of Large Language Models (LLMs), inspired by three-branch governmental systems. It implements three independent yet interacting components : LLMs as the executive branch for knowledge generation, Dike ...", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/380515639_Checks-and-Balances_Framework_for_Context-Aware_Ethical_AI_Alignment", "content": "This paper introduces a checks -and- balances framework for ethical alignment of Large Language Models (LLMs), inspired by three-branch governmental systems. It implements three independent yet interacting components : LLMs as the executive branch for knowledge generation, Dike ..."}
+{"idx": 1, "title": "A Checks -and- Balances Framework for Context-Aware Ethical AI ...", "date": "", "ddg_snippet": "This work introduces a checks -and- balances framework for ethical AI behavior. By delineating the responsibilities: LLM (executive), Dike (legislative), and Eris (judicial), the framework enables robust ethical oversight while...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2502.00136", "content": "This work introduces a checks -and- balances framework for ethical AI behavior. By delineating the responsibilities: LLM (executive), Dike (legislative), and Eris (judicial), the framework enables robust ethical oversight while..."}
+{"idx": 2, "title": "A Three-Branch Checks -and- Balances Framework", "date": "", "ddg_snippet": "This work presents a three-branch framework for ethical AI behavior, inspired by governmental checks and balances , centered on the DIKE - ERIS duality.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=o2afWIxjKD", "content": "This work presents a three-branch framework for ethical AI behavior, inspired by governmental checks and balances , centered on the DIKE - ERIS duality."}
+{"idx": 3, "title": "Ethical AI Alignment Frameworks | Restackio", "date": "", "ddg_snippet": "Ethical AI Alignment Frameworks . Last updated on 06/03/25.By aligning with the EU AI Act’s risk-based approach, AI systems use human opinions as feedback to adjust or pause the AI models when applied in high-risk scenarios (European Commission 2021, M. Veale 2021).", "subpage_snippet": "", "source": "d2wozrt205r2fu.cloudfront.net", "link": "https://d2wozrt205r2fu.cloudfront.net/p/ethical-ai-answer-alignment-frameworks-cat-ai", "content": "Ethical AI Alignment Frameworks . Last updated on 06/03/25.By aligning with the EU AI Act’s risk-based approach, AI systems use human opinions as feedback to adjust or pause the AI models when applied in high-risk scenarios (European Commission 2021, M. Veale 2021)."}
+{"idx": 4, "title": "Post-Human Value Alignment : Aligning with Values That Humans...", "date": "", "ddg_snippet": "This blog post delves into ethical frameworks , the challenges of defining post-human values, and the need for interdisciplinary collaboration to ensure that AI systems contribute positively to society.", "subpage_snippet": "", "source": "diversedaily.com", "link": "https://diversedaily.com/post-human-value-alignment-aligning-with-values-that-humans-might-never-conceive/", "content": "This blog post delves into ethical frameworks , the challenges of defining post-human values, and the need for interdisciplinary collaboration to ensure that AI systems contribute positively to society."}
+{"idx": 5, "title": "From Theory to Practice: Applying Ethical Frameworks to AI", "date": "", "ddg_snippet": "It covers key concepts such as ethical AI design, the AI alignment problem, and strategies for addressing ethical dilemmas in real-world scenarios.", "subpage_snippet": "", "source": "rutube.ru", "link": "https://rutube.ru/video/27b88aa32457ae889eff518f90c1f0d4/", "content": "It covers key concepts such as ethical AI design, the AI alignment problem, and strategies for addressing ethical dilemmas in real-world scenarios."}
+{"idx": 6, "title": "AI Alignment : Balancing Innovation with Ethical ... - Mahira", "date": "", "ddg_snippet": "The Challenge of AI Alignment . AI systems typically operate based on vast datasets and complex algorithms, making their decision-making processes difficult to interpret. This \"black box\" nature raises questions about accountability, transparency, and fairness.", "subpage_snippet": "", "source": "mahira.me", "link": "https://mahira.me/ai-alignment-balancing-innovation-with-ethical-considerations/", "content": "The Challenge of AI Alignment . AI systems typically operate based on vast datasets and complex algorithms, making their decision-making processes difficult to interpret. This \"black box\" nature raises questions about accountability, transparency, and fairness."}
+{"idx": 7, "title": "Navigating the Ethical Implications of AI in Digital Marketing...", "date": "", "ddg_snippet": "Investing in ethical AI training and awareness for marketing teams is another effective strategy. By educating employees about ethical considerations and the importance of consumer privacy, organizations can ensure that marketing practices align with established ethical standards.", "subpage_snippet": "", "source": "spreadbot.ai", "link": "https://spreadbot.ai/blog/navigating-the-ethical-implications-of-ai-in-digital-marketing-balancing-innovation-and-consumer-privacy/", "content": "Investing in ethical AI training and awareness for marketing teams is another effective strategy. By educating employees about ethical considerations and the importance of consumer privacy, organizations can ensure that marketing practices align with established ethical standards."}
+{"idx": 8, "title": "MACI: Multi-LLM Agent Collaborative Intelligence", "date": "", "ddg_snippet": "1. Checks and Balances for AI Ethical Alignment , as RLHF Fails. A Three-Branch Checks -and- Balances Framework for Context-Aware Ethical Alignment of Large Language Models Edward Y. Chang, NeurIPS AI Safety, December 2024.", "subpage_snippet": "", "source": "infolab.stanford.edu", "link": "http://infolab.stanford.edu/~echang/SocraSynth.html", "content": "1. Checks and Balances for AI Ethical Alignment , as RLHF Fails. A Three-Branch Checks -and- Balances Framework for Context-Aware Ethical Alignment of Large Language Models Edward Y. Chang, NeurIPS AI Safety, December 2024."}
+{"idx": 9, "title": "BLOG Post 9 of 10: Overview", "date": "", "ddg_snippet": "Figure 1: Ethical AI Framework for Data Privacy in Education This diagram illustrates the components of a robust AI ethics framework for data privacy in education, integrating UNESCO’s guidelines (UNESCO, 2022", "subpage_snippet": "", "source": "ai4edu.eu", "link": "https://ai4edu.eu/wp-content/uploads/2025/01/9.-Agentic-AI-Singularity-The-Integration-of-Artificial-Intelligence-in-Education-and-EdTech_AI-Collaboration-in-the-Workplace.pdf", "content": "Figure 1: Ethical AI Framework for Data Privacy in Education This diagram illustrates the components of a robust AI ethics framework for data privacy in education, integrating UNESCO’s guidelines (UNESCO, 2022"}
diff --git a/data/sampled_jsons/Disjunctive_counterfactuals_using_causal_models_a_critical_examination_Beckers_2023_abstract.jsonl b/data/sampled_jsons/Disjunctive_counterfactuals_using_causal_models_a_critical_examination_Beckers_2023_abstract.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..0268aff558a43f7bf1f717acf45cf90c8b764437
--- /dev/null
+++ b/data/sampled_jsons/Disjunctive_counterfactuals_using_causal_models_a_critical_examination_Beckers_2023_abstract.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Intervention and Conditioning in Causal Bayesian Networks", "date": "", "ddg_snippet": "by S Galhotra · 2024 · Cited by 3 — Beckers, S. (2023). Disjunctive counterfactuals using causal models: a critical examination . Un- published manuscript. Fariha, A., S. Nath, and A. Meliou ...", "subpage_snippet": "", "source": "proceedings.neurips.cc", "link": "https://proceedings.neurips.cc/paper_files/paper/2024/file/a2118322165fffb648d1e341ff5a5b05-Paper-Conference.pdf", "content": "by S Galhotra · 2024 · Cited by 3 — Beckers, S. (2023). Disjunctive counterfactuals using causal models: a critical examination . Un- published manuscript. Fariha, A., S. Nath, and A. Meliou ..."}
+{"idx": 1, "title": "Intervention and conditioning in causal Bayesian networks", "date": "", "ddg_snippet": "5 Jun 2025 — Beckers, S. (2023). Disjunctive counterfactuals using causal models: a critical examination . Unpublished manuscript. Google Scholar. [4].", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/10.5555/3737916.3740740", "content": "5 Jun 2025 — Beckers, S. (2023). Disjunctive counterfactuals using causal models: a critical examination . Unpublished manuscript. Google Scholar. [4]."}
+{"idx": 2, "title": "Causal Models with Constraints", "date": "", "ddg_snippet": "by S Beckers · 2023 · Cited by 9 — Abstract . Causal models have proven extremely useful in offering formal representations of causal relation- ships between a set of variables. 14 pages", "subpage_snippet": "", "source": "proceedings.mlr.press", "link": "https://proceedings.mlr.press/v213/beckers23a/beckers23a.pdf", "content": "by S Beckers · 2023 · Cited by 9 — Abstract . Causal models have proven extremely useful in offering formal representations of causal relation- ships between a set of variables. 14 pages"}
+{"idx": 3, "title": "Intervention and Conditioning in Causal Bayesian Networks", "date": "", "ddg_snippet": "by S Galhotra · 2024 · Cited by 3 — Beckers, S. (2023). Disjunctive counterfactuals using causal models: a critical examination . Un- published manuscript. Galhotra, S., R ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2405.14728?", "content": "by S Galhotra · 2024 · Cited by 3 — Beckers, S. (2023). Disjunctive counterfactuals using causal models: a critical examination . Un- published manuscript. Galhotra, S., R ..."}
+{"idx": 4, "title": "Intervention and Conditioning in Causal Bayesian Networks", "date": "", "ddg_snippet": "by J Halpern — Abstract : Causal models are crucial for understanding complex systems and identifying causal relationships among variables. Even though causal ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=DC28Fpk76s", "content": "by J Halpern — Abstract : Causal models are crucial for understanding complex systems and identifying causal relationships among variables. Even though causal ..."}
+{"idx": 5, "title": "Causal Modeling - Bibliography", "date": "", "ddg_snippet": "A causal model is a formal device intended to represent a part of the causal structure of the world. It comprises several variables and specifies how (and if) ...", "subpage_snippet": "", "source": "philpapers.org", "link": "https://philpapers.org/browse/causal-modeling", "content": "A causal model is a formal device intended to represent a part of the causal structure of the world. It comprises several variables and specifies how (and if) ..."}
+{"idx": 6, "title": "Reasoning About Causal Knowledge in Nondeterministic ...", "date": "", "ddg_snippet": "Abstract . Reasoning about causality and agent causal knowl- edge is critical for effective decision-making and planning in multi-agent contexts.", "subpage_snippet": "", "source": "www.ijcai.org", "link": "https://www.ijcai.org/proceedings/2025/0507.pdf", "content": "Abstract . Reasoning about causality and agent causal knowl- edge is critical for effective decision-making and planning in multi-agent contexts."}
+{"idx": 7, "title": "Moral Responsibility for AI Systems", "date": "", "ddg_snippet": "by S Beckers · 2023 · Cited by 6 — Beckers et. al. recently proposed a causal analysis of harm that is also formalized using causal models , and thus it could easily be integrated into the present ... 14 pages", "subpage_snippet": "", "source": "proceedings.neurips.cc", "link": "https://proceedings.neurips.cc/paper_files/paper/2023/file/0d5b7fd8c669fac58d6702188ed63afa-Paper-Conference.pdf", "content": "by S Beckers · 2023 · Cited by 6 — Beckers et. al. recently proposed a causal analysis of harm that is also formalized using causal models , and thus it could easily be integrated into the present ... 14 pages"}
+{"idx": 8, "title": "Moral Responsibility for AI Systems", "date": "", "ddg_snippet": "by S Beckers · 2023 · Cited by 6 — Beckers et. al. recently proposed a causal analysis of harm that is also formalized using causal models , and thus it could easily be integrated ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2310.18040", "content": "by S Beckers · 2023 · Cited by 6 — Beckers et. al. recently proposed a causal analysis of harm that is also formalized using causal models , and thus it could easily be integrated ..."}
+{"idx": 9, "title": "A Theoretical Foundation for Mechanistic Interpretability", "date": "", "ddg_snippet": "by A Geiger · 2025 · Cited by 53 — Abstract . Causal abstraction provides a theoretical foundation for mechanistic interpretability, the field concerned with providing intelligible algorithms ... 64 pages", "subpage_snippet": "", "source": "www.jmlr.org", "link": "https://www.jmlr.org/papers/volume26/23-0058/23-0058.pdf", "content": "by A Geiger · 2025 · Cited by 53 — Abstract . Causal abstraction provides a theoretical foundation for mechanistic interpretability, the field concerned with providing intelligible algorithms ... 64 pages"}
diff --git a/data/sampled_jsons/Enhanced_Language-Image_Toxicity_Evaluation_Safety_ELITE_equation_2_year_2024.jsonl b/data/sampled_jsons/Enhanced_Language-Image_Toxicity_Evaluation_Safety_ELITE_equation_2_year_2024.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..8fae30e13e41b16a92b1fa1ac16ce7e6ef724c10
--- /dev/null
+++ b/data/sampled_jsons/Enhanced_Language-Image_Toxicity_Evaluation_Safety_ELITE_equation_2_year_2024.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "velpegor/ ELITE : [ICML 2025] ELITE : Enhanced Language - Image ...", "date": "", "ddg_snippet": "Repository files navigation. README. ELITE : Enhanced Language - Image Toxicity Evaluation for Safety .", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/velpegor/ELITE", "content": "Repository files navigation. README. ELITE : Enhanced Language - Image Toxicity Evaluation for Safety ."}
+{"idx": 1, "title": "ELITE: Enhanced Language-Image Toxicity Evaluation for Safety How ELITE Reveals Dangerous Weaknesses in Vision-Language AI cvlab.yonsei.ac.kr ELITE: Enhanced Language-Image Toxicity Evaluation for Safety AIM Intelligence's ELITE Collaborative Paper Accepted by the ICML Doehyeon Lee | Seoul National University, Information ... ELITE: Enhanced Language-Image Toxicity Evaluation for Safety", "date": "", "ddg_snippet": "Feb 7, 2025 · Therefore, we found that existing benchmarks have low levels of harmfulness, ambiguous data, and limited diversity in image -text pair combinations. To address these issues, we propose the ELITE benchmark, a high-quality safety evaluation benchmark for VLMs, underpinned by our enhanced evaluation method, the ELITE evaluator. 2 . ELITE Evaluator: Grading Toxicity with Nuance The paper introduces a new evaluation formula based on the StrongREJECT rubric but adds a crucial factor: toxicity , which captures the degree of ... Therefore, we found that existing benchmarks have low levels of harmfulness, ambiguous data, and limited diversity in image -text pair combinations. To address these issues, we propose the ELITE benchmark, a high-quality safety evaluation benchmark for VLMs, underpinned by our enhanced evaluation method, the ELITE evaluator. The ELITE benchmark is proposed, a high-quality safety evaluation benchmark for VLMs, underpinned by the enhanced evaluation method, the ELITE evaluator, which explicitly incorporates a toxicity score to accurately assess harmfulness in multimodal contexts. Current Vision Language Models (VLMs) remain vulnerable to malicious prompts that induce harmful outputs. Existing safety benchmarks for ... May 15, 2025 · The paper proposes ELITE , a high-quality benchmark designed to evaluate the safety of Vision- Language Models (VLMs) with greater precision. At its core is the ELITE evaluator, a rubric-based method that incorporates a toxicity score to measure harmfulness in multimodal contexts-especially where VLMs produce specific, convincing responses that may appear harmless but convey dangerous intent. ELITE : Enhanced Language-Image Toxicity Evaluation for Safety Wonjun Lee*, Doehyeon Lee*, Eugene Choi, Sangyoon Yu, Ashkan Yousefpour, Haon Park, Bumsub Ham, Suhyun Kim (*Equal Contribution) Forty-second International Conference on Machine Learning, 2025 (ICML 2025) PDF Code Project Page BibTex Poster Presentation About [ICML 2025] ELITE : Enhanced Language-Image Toxicity Evaluation for Safety", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2502.04757", "content": "Feb 7, 2025 · Therefore, we found that existing benchmarks have low levels of harmfulness, ambiguous data, and limited diversity in image -text pair combinations. To address these issues, we propose the ELITE benchmark, a high-quality safety evaluation benchmark for VLMs, underpinned by our enhanced evaluation method, the ELITE evaluator. 2 . ELITE Evaluator: Grading Toxicity with Nuance The paper introduces a new evaluation formula based on the StrongREJECT rubric but adds a crucial factor: toxicity , which captures the degree of ... Therefore, we found that existing benchmarks have low levels of harmfulness, ambiguous data, and limited diversity in image -text pair combinations. To address these issues, we propose the ELITE benchmark, a high-quality safety evaluation benchmark for VLMs, underpinned by our enhanced evaluation method, the ELITE evaluator. The ELITE benchmark is proposed, a high-quality safety evaluation benchmark for VLMs, underpinned by the enhanced evaluation method, the ELITE evaluator, which explicitly incorporates a toxicity score to accurately assess harmfulness in multimodal contexts. Current Vision Language Models (VLMs) remain vulnerable to malicious prompts that induce harmful outputs. Existing safety benchmarks for ... May 15, 2025 · The paper proposes ELITE , a high-quality benchmark designed to evaluate the safety of Vision- Language Models (VLMs) with greater precision. At its core is the ELITE evaluator, a rubric-based method that incorporates a toxicity score to measure harmfulness in multimodal contexts-especially where VLMs produce specific, convincing responses that may appear harmless but convey dangerous intent. ELITE : Enhanced Language-Image Toxicity Evaluation for Safety Wonjun Lee*, Doehyeon Lee*, Eugene Choi, Sangyoon Yu, Ashkan Yousefpour, Haon Park, Bumsub Ham, Suhyun Kim (*Equal Contribution) Forty-second International Conference on Machine Learning, 2025 (ICML 2025) PDF Code Project Page BibTex Poster Presentation About [ICML 2025] ELITE : Enhanced Language-Image Toxicity Evaluation for Safety"}
+{"idx": 2, "title": "How ELITE Reveals Dangerous Weaknesses in Vision-Language AI", "date": "", "ddg_snippet": "2 . ELITE Evaluator: Grading Toxicity with Nuance The paper introduces a new evaluation formula based on the StrongREJECT rubric but adds a crucial factor: toxicity , which captures the degree of ...", "subpage_snippet": "", "source": "medium.com", "link": "https://medium.com/aim-intelligence/how-elite-reveals-dangerous-weaknesses-in-vision-language-ai-ffa208b7546c", "content": "2 . ELITE Evaluator: Grading Toxicity with Nuance The paper introduces a new evaluation formula based on the StrongREJECT rubric but adds a crucial factor: toxicity , which captures the degree of ..."}
+{"idx": 3, "title": "ELITE: Enhanced Language-Image Toxicity Evaluation for Safety", "date": "", "ddg_snippet": "The ELITE benchmark is proposed, a high-quality safety evaluation benchmark for VLMs, underpinned by the enhanced evaluation method, the ELITE evaluator, which explicitly incorporates a toxicity score to accurately assess harmfulness in multimodal contexts. Current Vision Language Models (VLMs) remain vulnerable to malicious prompts that induce harmful outputs. Existing safety benchmarks for ...", "subpage_snippet": "", "source": "www.semanticscholar.org", "link": "https://www.semanticscholar.org/paper/ELITE:-Enhanced-Language-Image-Toxicity-Evaluation-Lee-Lee/2bf4206276d5f574bbb2e13a56b29b4522fea675", "content": "The ELITE benchmark is proposed, a high-quality safety evaluation benchmark for VLMs, underpinned by the enhanced evaluation method, the ELITE evaluator, which explicitly incorporates a toxicity score to accurately assess harmfulness in multimodal contexts. Current Vision Language Models (VLMs) remain vulnerable to malicious prompts that induce harmful outputs. Existing safety benchmarks for ..."}
+{"idx": 4, "title": "AIM Intelligence's ELITE Collaborative Paper Accepted by the ICML", "date": "", "ddg_snippet": "May 15, 2025 · The paper proposes ELITE , a high-quality benchmark designed to evaluate the safety of Vision- Language Models (VLMs) with greater precision. At its core is the ELITE evaluator, a rubric-based method that incorporates a toxicity score to measure harmfulness in multimodal contexts-especially where VLMs produce specific, convincing responses that may appear harmless but convey dangerous intent.", "subpage_snippet": "", "source": "www.newsfilecorp.com", "link": "https://www.newsfilecorp.com/release/252268/AIM-Intelligences-ELITE-Collaborative-Paper-Accepted-by-the-ICML", "content": "May 15, 2025 · The paper proposes ELITE , a high-quality benchmark designed to evaluate the safety of Vision- Language Models (VLMs) with greater precision. At its core is the ELITE evaluator, a rubric-based method that incorporates a toxicity score to measure harmfulness in multimodal contexts-especially where VLMs produce specific, convincing responses that may appear harmless but convey dangerous intent."}
+{"idx": 5, "title": "ELITE : Enhanced Language - Image Toxicity Evaluation for Safety", "date": "", "ddg_snippet": "We identify these problems in the safety evaluation methods and propose the Enhanced Language - Image Toxicity Evaluation ( ELITE ) evaluator , a method designed to accurately evaluate the safety of VLMs.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2502.04757", "content": "We identify these problems in the safety evaluation methods and propose the Enhanced Language - Image Toxicity Evaluation ( ELITE ) evaluator , a method designed to accurately evaluate the safety of VLMs."}
+{"idx": 6, "title": "[Literature Review] ELITE : Enhanced Language - Image Toxicity ...", "date": "", "ddg_snippet": "The paper \" ELITE : Enhanced Language - Image Toxicity Evaluation for Safety \" presents a novel framework designed to evaluate the safety of Vision Language Models (VLMs) against harmful, malicious inputs that could generate unsafe content.", "subpage_snippet": "", "source": "www.themoonlight.io", "link": "https://www.themoonlight.io/en/review/elite-enhanced-language-image-toxicity-evaluation-for-safety", "content": "The paper \" ELITE : Enhanced Language - Image Toxicity Evaluation for Safety \" presents a novel framework designed to evaluate the safety of Vision Language Models (VLMs) against harmful, malicious inputs that could generate unsafe content."}
+{"idx": 7, "title": "Paper page - ELITE : Enhanced Language - Image Toxicity ...", "date": "", "ddg_snippet": "ELITE : Enhanced Language - Image Toxicity Evaluation for Safety .The ELITE evaluator explicitly incorporates a toxicity score to accurately assess harmfulness in multimodal contexts, where VLMs often provide specific, convincing, but unharmful descriptions of images.", "subpage_snippet": "", "source": "huggingface.co", "link": "https://huggingface.co/papers/2502.04757", "content": "ELITE : Enhanced Language - Image Toxicity Evaluation for Safety .The ELITE evaluator explicitly incorporates a toxicity score to accurately assess harmfulness in multimodal contexts, where VLMs often provide specific, convincing, but unharmful descriptions of images."}
+{"idx": 8, "title": "ELITE : Enhanced Language - Image Toxicity Evaluation for Safety", "date": "", "ddg_snippet": "Through the ELITE evaluator , we demonstrate that existing automated safety evaluation methods often result in inaccurate evaluations . • We propose the ELITE benchmark, a rubric-based safety evaluation benchmark for VLMs using the ELITE evaluator .", "subpage_snippet": "", "source": "bytez.com", "link": "https://bytez.com/docs/icml/46445/paper", "content": "Through the ELITE evaluator , we demonstrate that existing automated safety evaluation methods often result in inaccurate evaluations . • We propose the ELITE benchmark, a rubric-based safety evaluation benchmark for VLMs using the ELITE evaluator ."}
+{"idx": 9, "title": "AIM Intelligence's ELITE Collaborative Paper Accepted by the ICML", "date": "", "ddg_snippet": "AIM Intelligence CI. The paper proposes ELITE , a high-quality benchmark designed to evaluate the safety of Vision- Language Models (VLMs) with greater precision.", "subpage_snippet": "", "source": "www.digitaljournal.com", "link": "https://www.digitaljournal.com/pr/news/newsfile/aim-intelligence-s-elite-collaborative-paper-1773375521.html", "content": "AIM Intelligence CI. The paper proposes ELITE , a high-quality benchmark designed to evaluate the safety of Vision- Language Models (VLMs) with greater precision."}
diff --git a/data/sampled_jsons/Equation_(12)_convergence_rate_Likelihood_Based_Approach_to_Distribution_Regression.jsonl b/data/sampled_jsons/Equation_(12)_convergence_rate_Likelihood_Based_Approach_to_Distribution_Regression.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..c495fc009b2aca44dfc63707f688d0e382c9474d
--- /dev/null
+++ b/data/sampled_jsons/Equation_(12)_convergence_rate_Likelihood_Based_Approach_to_Distribution_Regression.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "A Likelihood Based Approach to Distribution Regression ...", "date": "", "ddg_snippet": "2 Oct 2024 — The convergence rate at ( 12 ) also recovers the optimal rate when q = 1 𝑞 1 q=1 italic_q = 1 and α = 0 𝛼 0 \\alpha=0 italic_α = 0 , and ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2410.02025v1", "content": "2 Oct 2024 — The convergence rate at ( 12 ) also recovers the optimal rate when q = 1 𝑞 1 q=1 italic_q = 1 and α = 0 𝛼 0 \\alpha=0 italic_α = 0 , and ..."}
+{"idx": 1, "title": "A Likelihood Based Approach to Distribution Regression Using...", "date": "", "ddg_snippet": "In this work, we explore the theoretical properties of conditional deep generative models under the statistical framework of distribution regression where ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=1IyPRv1A0r&noteId=LRN425LsSb", "content": "In this work, we explore the theoretical properties of conditional deep generative models under the statistical framework of distribution regression where ..."}
+{"idx": 2, "title": "A Likelihood Based Approach to Distribution Regression ...", "date": "", "ddg_snippet": "The convergence rate at ( 12 ) also recovers the optimal rate when q = 1 and α = 0, and there is a small lag of polynomial factor t∗α/(2β∗ + t∗) when α > 0 ...", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/poster/46645", "content": "The convergence rate at ( 12 ) also recovers the optimal rate when q = 1 and α = 0, and there is a small lag of polynomial factor t∗α/(2β∗ + t∗) when α > 0 ..."}
+{"idx": 3, "title": "A likelihood based approach to distribution regression ...", "date": "", "ddg_snippet": "by S Kumar · 2024 · Cited by 1 — A Likelihood Based Approach to Distribution Regression Using ... The convergence rate at ( 12 ) also recovers the optimal rate when q = 1 and ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2410.02025", "content": "by S Kumar · 2024 · Cited by 1 — A Likelihood Based Approach to Distribution Regression Using ... The convergence rate at ( 12 ) also recovers the optimal rate when q = 1 and ..."}
+{"idx": 4, "title": "Distribution-on-distribution regression with Wasserstein ...", "date": "", "ddg_snippet": "This result shows that our method achieves optimal convergence rates . That is, the convergence rates in Theorem 1 achieve the parametric rate n − 1 / 2 ...", "subpage_snippet": "", "source": "www.sciencedirect.com", "link": "https://www.sciencedirect.com/science/article/pii/S0047259X24000411", "content": "This result shows that our method achieves optimal convergence rates . That is, the convergence rates in Theorem 1 achieve the parametric rate n − 1 / 2 ..."}
+{"idx": 5, "title": "A Likelihood Approach to Nonparametric Estimation of a ...", "date": "", "ddg_snippet": "by M Chae · 2023 · Cited by 33 — We investigate statistical properties of a likelihood approach to nonparametric estimation of a singular distribution using deep generative models. 42 pages", "subpage_snippet": "", "source": "www.jmlr.org", "link": "https://www.jmlr.org/papers/volume24/21-1099/21-1099.pdf", "content": "by M Chae · 2023 · Cited by 33 — We investigate statistical properties of a likelihood approach to nonparametric estimation of a singular distribution using deep generative models. 42 pages"}
+{"idx": 6, "title": "A likelihood approach to nonparametric estimation of a ...", "date": "", "ddg_snippet": "1 Jan 2023 — We investigate statistical properties of a likelihood approach to nonparametric estimation of a singular distribution using deep generative ...", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/10.5555/3648699.3648776", "content": "1 Jan 2023 — We investigate statistical properties of a likelihood approach to nonparametric estimation of a singular distribution using deep generative ..."}
+{"idx": 7, "title": "Convergence Rates for Gaussian Mixtures of Experts", "date": "", "ddg_snippet": "by N Ho · 2022 · Cited by 66 — We provide a theoretical treatment of over-specified Gaussian mixtures of experts with covariate-free gating networks. We establish the convergence rates of ... 81 pages", "subpage_snippet": "", "source": "www.jmlr.org", "link": "https://www.jmlr.org/papers/volume23/20-1129/20-1129.pdf", "content": "by N Ho · 2022 · Cited by 66 — We provide a theoretical treatment of over-specified Gaussian mixtures of experts with covariate-free gating networks. We establish the convergence rates of ... 81 pages"}
+{"idx": 8, "title": "Refined Convergence Rates for Maximum Likelihood ...", "date": "", "ddg_snippet": "by T Manole · 2022 · Cited by 33 — Abstract. We revisit the classical problem of deriving con- vergence rates for the maximum likelihood es- timator (MLE) in finite mixture models. The.", "subpage_snippet": "", "source": "proceedings.mlr.press", "link": "https://proceedings.mlr.press/v162/manole22a/manole22a.pdf", "content": "by T Manole · 2022 · Cited by 33 — Abstract. We revisit the classical problem of deriving con- vergence rates for the maximum likelihood es- timator (MLE) in finite mixture models. The."}
+{"idx": 9, "title": "TIVE MODELS", "date": "", "ddg_snippet": "To the best of our knowledge, our study is the first attempt to explore the likelihood - based ap- proach for distributional regression using a conditional deep ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/notes/edits/attachment?id=bKSeeDBAan&name=pdf", "content": "To the best of our knowledge, our study is the first attempt to explore the likelihood - based ap- proach for distributional regression using a conditional deep ..."}
diff --git a/data/sampled_jsons/FD2_dataset_formula_Section_3_httpsarxiv.orghtml2410.02025v1.jsonl b/data/sampled_jsons/FD2_dataset_formula_Section_3_httpsarxiv.orghtml2410.02025v1.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..091906ed966d03d9ccc300be1e71f6fe240fa1a5
--- /dev/null
+++ b/data/sampled_jsons/FD2_dataset_formula_Section_3_httpsarxiv.orghtml2410.02025v1.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "GitHub - microsoft/ArxivFormula: This repo is used to release ... FD2 Object Detection Dataset by Technical vision arXiv.org e-Print archive 【论文阅读 | AAAI 2025 | FD2-Net：用于红外 - 可见光目标检测的频率... (aaai2025) FD2-Net: Frequency-Driven Feature Decomposition ... FD2-Net: Frequency-Driven Feature Decomposition Network for ...", "date": "", "ddg_snippet": "ArxivFormula is the first dataset framing mathematical formula detection as a joint task of formula entity detection and formula relation extraction, rather than a simple task of object detection or instance segmentation. It's constructed using a weak supervision approach and comprises 500K document images for training, 50K for validation and 50K f... See full list on github.com •We released annotations and origin document images of the ArxivFormula datasets (OneDrive), please refer to Get Data section . •We released some examples of the ArxivFormula datasets, please refer to ArxivFormula_examples.zip. See full list on github.com Most of existing mathematical formula detectors focus on detecting formula entities through object detection or instance segmentation techniques. However, these methods often fail to convey complete messages due to the absence of the contextual and layout information of mathematical formulas . For a more comprehensive understanding of mathematical f... See full list on github.com See full list on github.com This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https ://cla.opensource.microsoft.com. When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA. See full list on github.com This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-... See full list on github.com Connect Your Model With Program Logic Find utilities and guides to help you start using the FD2 project in your project. arXiv is a free distribution service and an open-access archive for nearly 2.4 million scholarly articles in the fields of physics, mathematics, computer science, quantitative biology, quantitative finance, statistics, electrical engineering and systems science, and economics. May 18, 2025 · 为解决这一问题，提出了一种新的用于 IVOD 的频率驱动特征分解网络 FD2 -Net，该网络能有效捕捉跨多模态视觉空间的互补信息所具有的独特频率表示。 Dec 24, 2024 · 为此，作者提出了Frequency-Driven Feature Decomposition Network （FD2Net），如下图所示，包括三个部分：特征分解编码器，多模态重建，多尺度检测头。 To solve this problem, we introduce a novel Frequency-Driven Feature Decomposition Network for IVOD, called FD2 -Net, which effectively captures the unique frequency representations of complementary information across multimodal visual spaces.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/microsoft/ArxivFormula", "content": "ArxivFormula is the first dataset framing mathematical formula detection as a joint task of formula entity detection and formula relation extraction, rather than a simple task of object detection or instance segmentation. It's constructed using a weak supervision approach and comprises 500K document images for training, 50K for validation and 50K f... See full list on github.com •We released annotations and origin document images of the ArxivFormula datasets (OneDrive), please refer to Get Data section . •We released some examples of the ArxivFormula datasets, please refer to ArxivFormula_examples.zip. See full list on github.com Most of existing mathematical formula detectors focus on detecting formula entities through object detection or instance segmentation techniques. However, these methods often fail to convey complete messages due to the absence of the contextual and layout information of mathematical formulas . For a more comprehensive understanding of mathematical f... See full list on github.com See full list on github.com This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https ://cla.opensource.microsoft.com. When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA. See full list on github.com This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-... See full list on github.com Connect Your Model With Program Logic Find utilities and guides to help you start using the FD2 project in your project. arXiv is a free distribution service and an open-access archive for nearly 2.4 million scholarly articles in the fields of physics, mathematics, computer science, quantitative biology, quantitative finance, statistics, electrical engineering and systems science, and economics. May 18, 2025 · 为解决这一问题，提出了一种新的用于 IVOD 的频率驱动特征分解网络 FD2 -Net，该网络能有效捕捉跨多模态视觉空间的互补信息所具有的独特频率表示。 Dec 24, 2024 · 为此，作者提出了Frequency-Driven Feature Decomposition Network （FD2Net），如下图所示，包括三个部分：特征分解编码器，多模态重建，多尺度检测头。 To solve this problem, we introduce a novel Frequency-Driven Feature Decomposition Network for IVOD, called FD2 -Net, which effectively captures the unique frequency representations of complementary information across multimodal visual spaces."}
+{"idx": 1, "title": "FD2 Object Detection Dataset by Technical vision", "date": "", "ddg_snippet": "Connect Your Model With Program Logic Find utilities and guides to help you start using the FD2 project in your project.", "subpage_snippet": "", "source": "universe.roboflow.com", "link": "https://universe.roboflow.com/technical-vision/fd2-uz0bi", "content": "Connect Your Model With Program Logic Find utilities and guides to help you start using the FD2 project in your project."}
+{"idx": 2, "title": "arXiv.org e-Print archive", "date": "", "ddg_snippet": "arXiv is a free distribution service and an open-access archive for nearly 2.4 million scholarly articles in the fields of physics, mathematics, computer science, quantitative biology, quantitative finance, statistics, electrical engineering and systems science, and economics.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/", "content": "arXiv is a free distribution service and an open-access archive for nearly 2.4 million scholarly articles in the fields of physics, mathematics, computer science, quantitative biology, quantitative finance, statistics, electrical engineering and systems science, and economics."}
+{"idx": 3, "title": "【论文阅读 | AAAI 2025 | FD2-Net：用于红外 - 可见光目标检测的频率...", "date": "", "ddg_snippet": "May 18, 2025 · 为解决这一问题，提出了一种新的用于 IVOD 的频率驱动特征分解网络 FD2 -Net，该网络能有效捕捉跨多模态视觉空间的互补信息所具有的独特频率表示。", "subpage_snippet": "", "source": "blog.csdn.net", "link": "https://blog.csdn.net/berling00/article/details/148045242", "content": "May 18, 2025 · 为解决这一问题，提出了一种新的用于 IVOD 的频率驱动特征分解网络 FD2 -Net，该网络能有效捕捉跨多模态视觉空间的互补信息所具有的独特频率表示。"}
+{"idx": 4, "title": "(aaai2025) FD2-Net: Frequency-Driven Feature Decomposition ...", "date": "", "ddg_snippet": "Dec 24, 2024 · 为此，作者提出了Frequency-Driven Feature Decomposition Network （FD2Net），如下图所示，包括三个部分：特征分解编码器，多模态重建，多尺度检测头。", "subpage_snippet": "", "source": "zhuanlan.zhihu.com", "link": "https://zhuanlan.zhihu.com/p/14485798000", "content": "Dec 24, 2024 · 为此，作者提出了Frequency-Driven Feature Decomposition Network （FD2Net），如下图所示，包括三个部分：特征分解编码器，多模态重建，多尺度检测头。"}
+{"idx": 5, "title": "FD2-Net: Frequency-Driven Feature Decomposition Network for ...", "date": "", "ddg_snippet": "To solve this problem, we introduce a novel Frequency-Driven Feature Decomposition Network for IVOD, called FD2 -Net, which effectively captures the unique frequency representations of complementary information across multimodal visual spaces.", "subpage_snippet": "", "source": "paperreading.club", "link": "https://paperreading.club/page?id=272246", "content": "To solve this problem, we introduce a novel Frequency-Driven Feature Decomposition Network for IVOD, called FD2 -Net, which effectively captures the unique frequency representations of complementary information across multimodal visual spaces."}
+{"idx": 6, "title": "Find Open Datasets and Machine Learning Projects | Kaggle", "date": "", "ddg_snippet": "Download Open Datasets on 1000s of Projects + Share Projects on One Platform. Explore Popular Topics Like Government, Sports, Medicine, Fintech, Food, More.", "subpage_snippet": "", "source": "www.kaggle.com", "link": "https://www.kaggle.com/datasets", "content": "Download Open Datasets on 1000s of Projects + Share Projects on One Platform. Explore Popular Topics Like Government, Sports, Medicine, Fintech, Food, More."}
+{"idx": 7, "title": "formula Keypoint Detection Dataset by Cattle", "date": "", "ddg_snippet": "221 open source animals images. formula dataset by Cattle.If you use this dataset in a research paper, please cite it using the following BibTeX: @misc{. formula - 3 oih 2 _ dataset", "subpage_snippet": "", "source": "universe.roboflow.com", "link": "https://universe.roboflow.com/cattle-erlpe/formula-3oih2", "content": "221 open source animals images. formula dataset by Cattle.If you use this dataset in a research paper, please cite it using the following BibTeX: @misc{. formula - 3 oih 2 _ dataset"}
+{"idx": 8, "title": "Free Public Datasets for Data Science Projects", "date": "", "ddg_snippet": "In this post we can find free public datasets for Data Science projects. There is a big number of datasets which cover different areas - machine learning, presentation, data analysis and visualization.", "subpage_snippet": "", "source": "datascientyst.com", "link": "https://datascientyst.com/datasets/", "content": "In this post we can find free public datasets for Data Science projects. There is a big number of datasets which cover different areas - machine learning, presentation, data analysis and visualization."}
+{"idx": 9, "title": "HTML -просмотрщик - Онлайн редактор и инструмент для...", "date": "", "ddg_snippet": "HTML -просмотрщик - мощный онлайн-инструмент для редактирования, предварительного просмотра и форматирования HTML -кода с подсветкой синтаксиса и тестированием адаптивного дизайна.Введите или вставьте ваш HTML в редактор.", "subpage_snippet": "", "source": "htmlonline.org", "link": "https://htmlonline.org/ru/", "content": "HTML -просмотрщик - мощный онлайн-инструмент для редактирования, предварительного просмотра и форматирования HTML -кода с подсветкой синтаксиса и тестированием адаптивного дизайна.Введите или вставьте ваш HTML в редактор."}
diff --git a/data/sampled_jsons/FlowDec-75m_DAC-75_SIGMOS_4.50_kbits.jsonl b/data/sampled_jsons/FlowDec-75m_DAC-75_SIGMOS_4.50_kbits.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..f13613e676497bbd78d47bacbf77c127b8d862aa
--- /dev/null
+++ b/data/sampled_jsons/FlowDec-75m_DAC-75_SIGMOS_4.50_kbits.jsonl
@@ -0,0 +1,3 @@
+{"idx": 0, "title": "FlowDec : A flow-based full-band general audio codec with high...", "date": "", "ddg_snippet": "FlowDec - 75 m : 75 Hz, multi-bitrate.14, we show spectrograms comparing FlowDec - 75 m and DAC - 75 on three examples with high harmonic content such as speech and isolated music instruments.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2503.01485v1", "content": "FlowDec - 75 m : 75 Hz, multi-bitrate.14, we show spectrograms comparing FlowDec - 75 m and DAC - 75 on three examples with high harmonic content such as speech and isolated music instruments."}
+{"idx": 1, "title": "FLOWDEC: A FLOW-BASED FULL-BAND GENERAL", "date": "", "ddg_snippet": "SIGMOS , and logSpecMSE. In SI-SDR and SIGMOS , ScoreDec, and ... Figure 15: Spectrograms comparing FlowDec-75m against DAC-75 as well as the initial decoder.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf/46581252e0bf80bf4efd8fcbc4002f8627f498bb.pdf", "content": "SIGMOS , and logSpecMSE. In SI-SDR and SIGMOS , ScoreDec, and ... Figure 15: Spectrograms comparing FlowDec-75m against DAC-75 as well as the initial decoder."}
+{"idx": 2, "title": "", "date": "", "ddg_snippet": "", "subpage_snippet": "", "source": "", "link": "", "content": ""}
diff --git a/data/sampled_jsons/FlowDec_paper_Table_8_SIGMOS_DAC-75.jsonl b/data/sampled_jsons/FlowDec_paper_Table_8_SIGMOS_DAC-75.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..798cbfb6d506fa796294eaf1b1cc7dd026f138fd
--- /dev/null
+++ b/data/sampled_jsons/FlowDec_paper_Table_8_SIGMOS_DAC-75.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "GitHub - facebookresearch/FlowDec: An neural full-band audio codec for ...", "date": "", "ddg_snippet": "Frequency-dependent sigma_y For automatically determining the frequency-dependent sigma_y (see Section 3.5 in our paper ), you can use the helper script scripts/estimate_flowdec_params.py. This script also implements the heuristic for a global sigma_y discussed in our Appendix A.1.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/facebookresearch/FlowDec", "content": "Frequency-dependent sigma_y For automatically determining the frequency-dependent sigma_y (see Section 3.5 in our paper ), you can use the helper script scripts/estimate_flowdec_params.py. This script also implements the heuristic for a global sigma_y discussed in our Appendix A.1."}
+{"idx": 1, "title": "FlowDec: A flow-based full-band general audio codec with high...", "date": "", "ddg_snippet": "This paper proposes FlowDec , a 48 kHz general audio codec with a flow-matching diffusion post-filter. FlowDec modifies the DAC audio codec with different loss functions, the stochastic post-filter, and frequency-dependent noise.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=uxDFlPGRLX", "content": "This paper proposes FlowDec , a 48 kHz general audio codec with a flow-matching diffusion post-filter. FlowDec modifies the DAC audio codec with different loss functions, the stochastic post-filter, and frequency-dependent noise."}
+{"idx": 2, "title": "FlowDec: A flow-based full-band general audio codec with high ...", "date": "", "ddg_snippet": "Below we present audio examples from a low-feature-rate variant of FlowDec ( FlowDec -25s) and a retrained DAC ( DAC -25). These operate at a feature rate of 25 Hz instead of around 75 Hz, and are aimed at generative audio tasks where low feature rates are desirable for long-range sequence modeling.", "subpage_snippet": "", "source": "sp-uhh.github.io", "link": "https://sp-uhh.github.io/FlowDec/", "content": "Below we present audio examples from a low-feature-rate variant of FlowDec ( FlowDec -25s) and a retrained DAC ( DAC -25). These operate at a feature rate of 25 Hz instead of around 75 Hz, and are aimed at generative audio tasks where low feature rates are desirable for long-range sequence modeling."}
+{"idx": 3, "title": "FlowDec: A flow-based full-band general audio codec with high ...", "date": "", "ddg_snippet": "We propose FlowDec , a neural full-band audio codec for general audio sampled at 48 kHz that combines non-adversarial codec training with a stochastic postfilter based on a novel conditional flow matching method. Compared to the prior work ScoreDec which is based on score matching, we generalize from speech to general audio and move from 24 kbit/s to as low as 4 kbit/s, while improving output ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2503.01485", "content": "We propose FlowDec , a neural full-band audio codec for general audio sampled at 48 kHz that combines non-adversarial codec training with a stochastic postfilter based on a novel conditional flow matching method. Compared to the prior work ScoreDec which is based on score matching, we generalize from speech to general audio and move from 24 kbit/s to as low as 4 kbit/s, while improving output ..."}
+{"idx": 4, "title": "Releases · facebookresearch/FlowDec - GitHub", "date": "", "ddg_snippet": "The release includes both NDAC and FlowDec -postfilter checkpoints. The folder structure is as follows. checkpoints flowdec flowdec_25s flowdec_75m ndac ndac-25 ndac- 75", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/facebookresearch/FlowDec/releases", "content": "The release includes both NDAC and FlowDec -postfilter checkpoints. The folder structure is as follows. checkpoints flowdec flowdec_25s flowdec_75m ndac ndac-25 ndac- 75"}
+{"idx": 5, "title": "FlowDec: A flow-based full-band general audio codec with high ...", "date": "", "ddg_snippet": "We show that FlowDec is a competitive alternative to the recent GAN-dominated stream of neural codecs, achieving FAD scores better than those of the established GAN-based codec DAC and listening test scores that are on par, and producing qualitatively more natural reconstructions for speech and harmonic structures in music.", "subpage_snippet": "", "source": "www.alphaxiv.org", "link": "https://www.alphaxiv.org/overview/2503.01485", "content": "We show that FlowDec is a competitive alternative to the recent GAN-dominated stream of neural codecs, achieving FAD scores better than those of the established GAN-based codec DAC and listening test scores that are on par, and producing qualitatively more natural reconstructions for speech and harmonic structures in music."}
+{"idx": 6, "title": "FlowDec:An neural full-band audio codec for general audio sampled at 48 ...", "date": "", "ddg_snippet": "FlowDec FlowDec (ICLR 2025) is a full-band audio codec for general audio sampled at 48 kHz that combines non-adversarial codec training with a stochastic postfilter based on a novel conditional flow matching method.", "subpage_snippet": "", "source": "gitcode.com", "link": "https://gitcode.com/gh_mirrors/fl/FlowDec/overview", "content": "FlowDec FlowDec (ICLR 2025) is a full-band audio codec for general audio sampled at 48 kHz that combines non-adversarial codec training with a stochastic postfilter based on a novel conditional flow matching method."}
+{"idx": 7, "title": "FlowDec: A flow-based full-band general audio codec with high ...", "date": "", "ddg_snippet": "These results suggest that FlowDec may perform better on speech than DAC , particularly for FlowDec -75m versus DAC-75 at 4.5 kbit/s and that DAC may perform slightly better than FlowDec on sound files; score distributions for music are very similar.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2503.01485v1", "content": "These results suggest that FlowDec may perform better on speech than DAC , particularly for FlowDec -75m versus DAC-75 at 4.5 kbit/s and that DAC may perform slightly better than FlowDec on sound files; score distributions for music are very similar."}
+{"idx": 8, "title": "FlowDec: A flow-based full-band general audio codec with high ...", "date": "", "ddg_snippet": "We propose FlowDec , a neural full-band audio codec for general audio sampled at 48 kHz that combines non-adversarial codec training with a stochastic postfilter based on a novel conditional flow ...", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/389580542_FlowDec_A_flow-based_full-band_general_audio_codec_with_high_perceptual_quality", "content": "We propose FlowDec , a neural full-band audio codec for general audio sampled at 48 kHz that combines non-adversarial codec training with a stochastic postfilter based on a novel conditional flow ..."}
+{"idx": 9, "title": "A arXiv:2503.01485v1 [cs.SD] 3 Mar 2025", "date": "", "ddg_snippet": "In Fig. 4, we show the objective metric results of FlowDec -75m and FlowDec - 75s compared to EnCodec (48kHz), DAC-75 , 2xDAC- 75 and the official DAC 44.1kHz checkpoint, and also include 8", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2503.01485v1", "content": "In Fig. 4, we show the objective metric results of FlowDec -75m and FlowDec - 75s compared to EnCodec (48kHz), DAC-75 , 2xDAC- 75 and the official DAC 44.1kHz checkpoint, and also include 8"}
diff --git a/data/sampled_jsons/Hierarchical_Overlapping_Clustering_on_Graphs_Cost_Function_Algorithm_and_Scalability_time_complexit.jsonl b/data/sampled_jsons/Hierarchical_Overlapping_Clustering_on_Graphs_Cost_Function_Algorithm_and_Scalability_time_complexit.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..07c06f99546c7f3c5f8d724df3a956336846ea8b
--- /dev/null
+++ b/data/sampled_jsons/Hierarchical_Overlapping_Clustering_on_Graphs_Cost_Function_Algorithm_and_Scalability_time_complexit.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Hierarchical Overlapping Clustering Function Algorithm and Scalability ...", "date": "", "ddg_snippet": "Overlap and hierarchy are two prevalent phenomena in clustering , and usually coexist in a single system. There are several studies on each of them separately, but it is unclear how to characterize and evaluate the hybrid structures yet. To address this issue, we initiate the study of hierarchical overlapping clustering on graphs by introducing a new cost function for it. We show the ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=oHSXRy29tj", "content": "Overlap and hierarchy are two prevalent phenomena in clustering , and usually coexist in a single system. There are several studies on each of them separately, but it is unclear how to characterize and evaluate the hybrid structures yet. To address this issue, we initiate the study of hierarchical overlapping clustering on graphs by introducing a new cost function for it. We show the ..."}
+{"idx": 1, "title": "PDF Overlapping Hierarchical Clustering (OHC) - inria.hal.science", "date": "", "ddg_snippet": "In this paper we propose a new method that allows clusters to overlap until a strong cluster attraction is reached, based on a density criterion. The resulting hierarchical structure, called a quasi-dendrogram, is represented as a directed acyclic graph and com-bines the advantages of hierarchies with the precision of a less arbitrary clustering .", "subpage_snippet": "", "source": "inria.hal.science", "link": "https://inria.hal.science/hal-02452729/file/Overlapping_Hierarchical_Clustering_IDA2020_Camera_Ready_.pdf", "content": "In this paper we propose a new method that allows clusters to overlap until a strong cluster attraction is reached, based on a density criterion. The resulting hierarchical structure, called a quasi-dendrogram, is represented as a directed acyclic graph and com-bines the advantages of hierarchies with the precision of a less arbitrary clustering ."}
+{"idx": 2, "title": "Hierarchical Overlapping Clustering on Graphs: Cost Function, Algorithm ...", "date": "", "ddg_snippet": "This research paper introduces a new way to group data points in a more complex and realistic manner called hierarchical overlapping clustering ( HOC ). It combines two methods: hierarchical cluster...", "subpage_snippet": "", "source": "bytez.com", "link": "https://bytez.com/docs/icml/46447/paper", "content": "This research paper introduces a new way to group data points in a more complex and realistic manner called hierarchical overlapping clustering ( HOC ). It combines two methods: hierarchical cluster..."}
+{"idx": 3, "title": "Hierarchical Overlapping Clustering on Graphs: Cost Function, Algorithm ...", "date": "", "ddg_snippet": "Poster Hierarchical Overlapping Clustering on Graphs : Cost Function , Algorithm and Scalability Yicheng Pan · Renjie Chen · Pengyu Long · Bingchen Fan East Exhibition Hall A-B #E-2009", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/poster/46447", "content": "Poster Hierarchical Overlapping Clustering on Graphs : Cost Function , Algorithm and Scalability Yicheng Pan · Renjie Chen · Pengyu Long · Bingchen Fan East Exhibition Hall A-B #E-2009"}
+{"idx": 4, "title": "PDF Nearly-Optimal Hierarchical Clustering for Well-Clustered Graphs", "date": "", "ddg_snippet": "Abstract This paper presents two eficient hierarchical clus-tering (HC) algorithms with respect to Dasgupta's cost function . For any input graph G with a clear cluster-structure, our designed algorithms run in nearly-linear time in the input size of G, and re-turn an O(1)-approximate HC tree with respect to Dasgupta's cost function . We compare the perfor-mance of our algorithm against the ...", "subpage_snippet": "", "source": "proceedings.mlr.press", "link": "https://proceedings.mlr.press/v202/laenen23a/laenen23a.pdf", "content": "Abstract This paper presents two eficient hierarchical clus-tering (HC) algorithms with respect to Dasgupta's cost function . For any input graph G with a clear cluster-structure, our designed algorithms run in nearly-linear time in the input size of G, and re-turn an O(1)-approximate HC tree with respect to Dasgupta's cost function . We compare the perfor-mance of our algorithm against the ..."}
+{"idx": 5, "title": "Hierarchical Clustering: Objective Functions and Algorithms", "date": "", "ddg_snippet": "Motivated by the fact that most work on hierarchical clustering was based on providing algorithms , rather than optimizing a specific objective, Dasgupta framed similarity-based hierarchical clustering as a combinatorial optimization problem, where a \"good\" hierarchical clustering is one that minimizes a particular cost function [23].", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/fullHtml/10.1145/3321386", "content": "Motivated by the fact that most work on hierarchical clustering was based on providing algorithms , rather than optimizing a specific objective, Dasgupta framed similarity-based hierarchical clustering as a combinatorial optimization problem, where a \"good\" hierarchical clustering is one that minimizes a particular cost function [23]."}
+{"idx": 6, "title": "Hierarchical Clustering: $O(1)$-Approximation for Well-Clustered Graphs", "date": "", "ddg_snippet": "In this work we study the cost function for hierarchical clustering introduced by Dasgupta, and present two polynomial- time approximation algorithms : Our first result is an O(1) -approximation algorithm for graphs of high conductance. Our simple construction bypasses complicated recursive routines of finding sparse cuts known in the literature.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2112.09055", "content": "In this work we study the cost function for hierarchical clustering introduced by Dasgupta, and present two polynomial- time approximation algorithms : Our first result is an O(1) -approximation algorithm for graphs of high conductance. Our simple construction bypasses complicated recursive routines of finding sparse cuts known in the literature."}
+{"idx": 7, "title": "Nearly-Optimal Hierarchical Clustering for Well-Clustered Graphs", "date": "", "ddg_snippet": "Abstract This paper presents two efficient hierarchical clustering (HC) algorithms with respect to Dasgupta's cost function .", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/371684528_Nearly-Optimal_Hierarchical_Clustering_for_Well-Clustered_Graphs", "content": "Abstract This paper presents two efficient hierarchical clustering (HC) algorithms with respect to Dasgupta's cost function ."}
+{"idx": 8, "title": "Hierarchical Overlapping Clustering on Graphs: Cost Function, Algorithm ...", "date": "", "ddg_snippet": "To bridge this gap, we initiate the study of hierarchical overlapping clustering on graphs by introducing a new cost function and establishing its rationality through several intu-itive properties. We further develop an approxi-mation algorithm that achieves a constant approx-imation factor for its dual version.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=51x0dfsD8A", "content": "To bridge this gap, we initiate the study of hierarchical overlapping clustering on graphs by introducing a new cost function and establishing its rationality through several intu-itive properties. We further develop an approxi-mation algorithm that achieves a constant approx-imation factor for its dual version."}
+{"idx": 9, "title": "PDF G O arXiv:2306.09950v1 [cs.DS] 16 Jun 2023", "date": "", "ddg_snippet": "This paper presents two eficient hierarchical clustering (HC) algorithms with respect to Dasgupta's cost function . For any input graph G with a clear cluster-structure, our designed algorithms run in nearly-linear time in the input size of G, and return an O(1)-approximate HC tree with respect to Dasgupta's cost function . We compare the performance of our algorithm against the previous ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2306.09950.pdf", "content": "This paper presents two eficient hierarchical clustering (HC) algorithms with respect to Dasgupta's cost function . For any input graph G with a clear cluster-structure, our designed algorithms run in nearly-linear time in the input size of G, and return an O(1)-approximate HC tree with respect to Dasgupta's cost function . We compare the performance of our algorithm against the previous ..."}
diff --git a/data/sampled_jsons/HtmlRAG_Section_3.2.2_Lossless_Structural_Compression_single-nested_tags_year_2024.jsonl b/data/sampled_jsons/HtmlRAG_Section_3.2.2_Lossless_Structural_Compression_single-nested_tags_year_2024.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..d9c5cbdbd4a9a9b322d77836d6cf7750c53ea5f3
--- /dev/null
+++ b/data/sampled_jsons/HtmlRAG_Section_3.2.2_Lossless_Structural_Compression_single-nested_tags_year_2024.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "HtmlRAG : HTML is Better Than Plain Text for Modeling Retrieved...", "date": "", "ddg_snippet": "3 . 2 . 2 . Lossless Structural Compression .After block deleting, redundant HTML structures will re-appear, so we re-adjust the HTML structure , meaning multiple layers of single - nested tags are merged and empty tags are removed.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2411.02959v2", "content": "3 . 2 . 2 . Lossless Structural Compression .After block deleting, redundant HTML structures will re-appear, so we re-adjust the HTML structure , meaning multiple layers of single - nested tags are merged and empty tags are removed."}
+{"idx": 1, "title": "GitHub - plageon/ HtmlRAG : HtmlRAG : HTML is Better Than Plain...", "date": "", "ddg_snippet": "The compressed HTML of lossless HTML cleaning is suitable for RAG systems that have long-context LLMs and are not willing to loss any information before generation. Two-Step Block-Tree-Based HTML Pruning: The block-tree-based HTML pruning consists of two steps, both of...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/plageon/HtmlRAG", "content": "The compressed HTML of lossless HTML cleaning is suitable for RAG systems that have long-context LLMs and are not willing to loss any information before generation. Two-Step Block-Tree-Based HTML Pruning: The block-tree-based HTML pruning consists of two steps, both of..."}
+{"idx": 2, "title": "HtmlRAG : HTML is Better Than Plain Text for Modeling Retrieved...", "date": "", "ddg_snippet": "Lossless Structure Compression . ④Generative Fine structure , meaning multiple layers of single - nested tags are merged.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=E91gjsccP1", "content": "Lossless Structure Compression . ④Generative Fine structure , meaning multiple layers of single - nested tags are merged."}
+{"idx": 3, "title": "Implementing HtmlRAG : Enhancing Retrieval-Augmented... | Dev Genius", "date": "", "ddg_snippet": "• Compress Structure : • Merge nested tags where possible to reduce depth.6. Conclusion. Implementing HtmlRAG enhances RAG systems by utilizing the rich structural and semantic information present in HTML documents.", "subpage_snippet": "", "source": "blog.devgenius.io", "link": "https://blog.devgenius.io/implementing-htmlrag-enhancing-retrieval-augmented-generation-with-html-knowledge-91cdd6278e23", "content": "• Compress Structure : • Merge nested tags where possible to reduce depth.6. Conclusion. Implementing HtmlRAG enhances RAG systems by utilizing the rich structural and semantic information present in HTML documents."}
+{"idx": 4, "title": "HtmlRAG : HTML is Better Than Plain Text for Modeling Retrieved...", "date": "", "ddg_snippet": "The cleaning process also includes structural compression techniques such as merging multiple layers of nested tags and removing empty tags .", "subpage_snippet": "", "source": "deep-diver.github.io", "link": "https://deep-diver.github.io/ai-paper-reviewer/paper-reviews/2411.02959/", "content": "The cleaning process also includes structural compression techniques such as merging multiple layers of nested tags and removing empty tags ."}
+{"idx": 5, "title": "ℹ 1 0 1 **What is HtmlRAG , Multimodal RAG and Agentic...", "date": "", "ddg_snippet": "HtmlRAG works directly with HTML to keep more of the structure and meaning of the original content, missing the step of converting the data into the plain text. The main parts of HtmlRAG ’s working process are HTML Cleaning and Pruning techniques", "subpage_snippet": "", "source": "huggingface.co", "link": "https://huggingface.co/blog/Kseniase/html-multimodal-agentic-rag", "content": "HtmlRAG works directly with HTML to keep more of the structure and meaning of the original content, missing the step of converting the data into the plain text. The main parts of HtmlRAG ’s working process are HTML Cleaning and Pruning techniques"}
+{"idx": 6, "title": "HtmlRAG : Enhancing RAG Systems with Richer... - MarkTechPost", "date": "", "ddg_snippet": "HtmlRAG implements a two-step pruning mechanism to process retrieved HTML documents efficiently. Initially, the system concatenates all retrieved HTML documents and parses them into a single DOM tree using Beautiful Soup.", "subpage_snippet": "", "source": "www.marktechpost.com", "link": "https://www.marktechpost.com/2024/11/10/htmlrag-enhancing-rag-systems-with-richer-semantic-and-structural-information-through-html/", "content": "HtmlRAG implements a two-step pruning mechanism to process retrieved HTML documents efficiently. Initially, the system concatenates all retrieved HTML documents and parses them into a single DOM tree using Beautiful Soup."}
+{"idx": 7, "title": "(PDF) HtmlRAG : HTML is Better Than Plain Text for Modeling...", "date": "", "ddg_snippet": "Cleaning. Lossless Structure . Compression .dant HTML structures will re-appear, so we re-adjust the HTML . structure , meaning multiple layers of single - nested tags are merged. and empty tags are removed. The detailed pruning algorithm is.", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/385560345_HtmlRAG_HTML_is_Better_Than_Plain_Text_for_Modeling_Retrieved_Knowledge_in_RAG_Systems", "content": "Cleaning. Lossless Structure . Compression .dant HTML structures will re-appear, so we re-adjust the HTML . structure , meaning multiple layers of single - nested tags are merged. and empty tags are removed. The detailed pruning algorithm is."}
+{"idx": 8, "title": "plageon/ HtmlRAG | DeepWiki", "date": "", "ddg_snippet": "HtmlRAG is a system that leverages HTML structure rather than plain text for Retrieval-Augmented Generation ( RAG ) systems. This document provides a comprehensive overview of the HtmlRAG repository, in.", "subpage_snippet": "", "source": "deepwiki.com", "link": "https://deepwiki.com/plageon/HtmlRAG", "content": "HtmlRAG is a system that leverages HTML structure rather than plain text for Retrieval-Augmented Generation ( RAG ) systems. This document provides a comprehensive overview of the HtmlRAG repository, in."}
+{"idx": 9, "title": "HtmlRAG : Building an Efficient HTML Retrieval Enhanced Generation...", "date": "", "ddg_snippet": "Lossless HTML cleanup compressed HTML is suitable for RAG systems with long context LLMs and an unwillingness to lose any information before generation.", "subpage_snippet": "", "source": "www.kdjingpai.com", "link": "https://www.kdjingpai.com/en/htmlrag/", "content": "Lossless HTML cleanup compressed HTML is suitable for RAG systems with long context LLMs and an unwillingness to lose any information before generation."}
diff --git a/data/sampled_jsons/I(p)_=_Fisher_information_Remark_4.2_regularized_unbalanced_optimal_transport.jsonl b/data/sampled_jsons/I(p)_=_Fisher_information_Remark_4.2_regularized_unbalanced_optimal_transport.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..ea381d86ce8d8753aedfa129f77843c484e436d7
--- /dev/null
+++ b/data/sampled_jsons/I(p)_=_Fisher_information_Remark_4.2_regularized_unbalanced_optimal_transport.jsonl
@@ -0,0 +1,5 @@
+{"idx": 0, "title": "Learning stochastic dynamics from snapshots through regularized ...", "date": "", "ddg_snippet": "Unbalanced Dynamic Optimal Transport . Fisher information regularization schemes for wasserstein gradient flows.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2410.00844v4", "content": "Unbalanced Dynamic Optimal Transport . Fisher information regularization schemes for wasserstein gradient flows."}
+{"idx": 1, "title": "Learning stochastic dynamics from snapshots through regularized ...", "date": "", "ddg_snippet": "Unbalanced Dynamic Optimal Transport . Fisher information regularization schemes for wasserstein gradient flows.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2410.00844v5", "content": "Unbalanced Dynamic Optimal Transport . Fisher information regularization schemes for wasserstein gradient flows."}
+{"idx": 2, "title": "Learning stochastic dynamics from snapshots through ...", "date": "", "ddg_snippet": "Here, we introduce a new deep learning approach for solving regularized unbalanced optimal transport (RUOT) and inferring continuous unbalanced stochastic dynamics from observed snapshots.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2410.00844v3", "content": "Here, we introduce a new deep learning approach for solving regularized unbalanced optimal transport (RUOT) and inferring continuous unbalanced stochastic dynamics from observed snapshots."}
+{"idx": 3, "title": "Learning stochastic dynamics from snapshots through ...", "date": "", "ddg_snippet": "Abstract Reconstructing dynamics using samples from sparsely time-resolved snapshots is an important problem in both natural sciences and machine learning. Here, we introduce a new deep learning approach for solving regularized unbalanced optimal transport (RUOT) and inferring continuous unbalanced stochastic dynamics from observed snapshots.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2410.00844v2", "content": "Abstract Reconstructing dynamics using samples from sparsely time-resolved snapshots is an important problem in both natural sciences and machine learning. Here, we introduce a new deep learning approach for solving regularized unbalanced optimal transport (RUOT) and inferring continuous unbalanced stochastic dynamics from observed snapshots."}
+{"idx": 4, "title": "LEARNING STOCHASTIC DYNAMICS FROM SNAP SHOTS THROUGH ...", "date": "", "ddg_snippet": "ABSTRACT Reconstructing dynamics using samples from sparsely time-resolved snapshots is an important problem in both natural sciences and machine learning. Here, we introduce a new deep learning approach for solving regularized unbalanced optimal transport (RUOT) and inferring continuous unbalanced stochastic dy- namics from observed snapshots.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=gQlxd3Mtru", "content": "ABSTRACT Reconstructing dynamics using samples from sparsely time-resolved snapshots is an important problem in both natural sciences and machine learning. Here, we introduce a new deep learning approach for solving regularized unbalanced optimal transport (RUOT) and inferring continuous unbalanced stochastic dy- namics from observed snapshots."}
diff --git a/data/sampled_jsons/In_ERL,_policies_predict_entire_action_trajectories_over_multiple_time_steps_instead_of_single_per-s.jsonl b/data/sampled_jsons/In_ERL,_policies_predict_entire_action_trajectories_over_multiple_time_steps_instead_of_single_per-s.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..0ae870369f2ff5599ffb4917b7d0d46ec8226e26
--- /dev/null
+++ b/data/sampled_jsons/In_ERL,_policies_predict_entire_action_trajectories_over_multiple_time_steps_instead_of_single_per-s.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Coordinated management of meaning - Wikipedia", "date": "", "ddg_snippet": "Assuming that persons transform sensory perceptions into implications for meaning and action , and that of the process for this transformation may be ...", "subpage_snippet": "", "source": "en.wikipedia.org", "link": "https://en.wikipedia.org/wiki/Coordinated_management_of_meaning", "content": "Assuming that persons transform sensory perceptions into implications for meaning and action , and that of the process for this transformation may be ..."}
+{"idx": 1, "title": "TOP-ERL: TRANSFORMER BASED OFF-POLICY E REINFORCEMENT ...", "date": "", "ddg_snippet": "s over multiple time steps instead of single per - step actions . These trajectories are typically parame-terized by trajectory generators such as Movement Primitives (MP), allowing for smooth and eficient explora.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2410.09536v2", "content": "s over multiple time steps instead of single per - step actions . These trajectories are typically parame-terized by trajectory generators such as Movement Primitives (MP), allowing for smooth and eficient explora."}
+{"idx": 2, "title": "TOP-ERL: Transformer-based Off-Policy Episodic Reinforcement ...", "date": "", "ddg_snippet": "Abstract—This work introduces Transformer-based Off-Policy Episodic Reinforcement Learning ( TOP - ERL ), a novel algorithm that enables off-policy updates in an ERL framework. In ERL , policies predict entire action trajectories over multiple time steps instead of single per - step actions .", "subpage_snippet": "", "source": "rss25-roboreps.github.io", "link": "https://rss25-roboreps.github.io/papers/1_TOP_ERL_Transformer_based_Of.pdf", "content": "Abstract—This work introduces Transformer-based Off-Policy Episodic Reinforcement Learning ( TOP - ERL ), a novel algorithm that enables off-policy updates in an ERL framework. In ERL , policies predict entire action trajectories over multiple time steps instead of single per - step actions ."}
+{"idx": 3, "title": "RANKED! The 100 best individual performances in football |", "date": "", "ddg_snippet": "Usually deployed as a centre-back who would step into midfield, Stones instead reverted to his early-career right-back role out of possession.", "subpage_snippet": "", "source": "www.fourfourtwo.com", "link": "https://www.fourfourtwo.com/features/ranked-the-100-best-individual-performances-in-football-ever", "content": "Usually deployed as a centre-back who would step into midfield, Stones instead reverted to his early-career right-back role out of possession."}
+{"idx": 4, "title": "E pisodic r einforcement L earning", "date": "", "ddg_snippet": "This work introduces Transformer-based Off-Policy Episodic Reinforcement Learning ( TOP - ERL ), a novel algorithm that enables off-policy updates in an ERL framework. In ERL , policies predict entire action trajectories over multiple time steps instead of single per - step actions .", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2410.09536", "content": "This work introduces Transformer-based Off-Policy Episodic Reinforcement Learning ( TOP - ERL ), a novel algorithm that enables off-policy updates in an ERL framework. In ERL , policies predict entire action trajectories over multiple time steps instead of single per - step actions ."}
+{"idx": 5, "title": "GRC 2025 Program | Thursday March 13, 2025", "date": "", "ddg_snippet": "In ERL , policies predict entire action trajectories over multiple time steps instead of single per - step actions .", "subpage_snippet": "", "source": "ras.papercept.net", "link": "https://ras.papercept.net/conferences/conferences/GRC25/program/GRC25_ContentListWeb_1.html", "content": "In ERL , policies predict entire action trajectories over multiple time steps instead of single per - step actions ."}
+{"idx": 6, "title": "Implicit Reasoning in Large Language Models: A Comprehensive", "date": "", "ddg_snippet": "... of -thought prompting toward implicit reasoning, where reasoning occurs silently via latent structures without emitting intermediate textual steps ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2509.02350v1", "content": "... of -thought prompting toward implicit reasoning, where reasoning occurs silently via latent structures without emitting intermediate textual steps ..."}
+{"idx": 7, "title": "DYSPEPSIA GENERATION » 2025 » May", "date": "", "ddg_snippet": "Trump Officials Unveil Budget Cuts to Aid for Health, Housing and Research (Tony Romm/New York Times ) Because health, housing, and research are ...", "subpage_snippet": "", "source": "dyspepsiageneration.com", "link": "https://dyspepsiageneration.com/?m=202505", "content": "Trump Officials Unveil Budget Cuts to Aid for Health, Housing and Research (Tony Romm/New York Times ) Because health, housing, and research are ..."}
+{"idx": 8, "title": "- FISO - The UK's First Fantasy Sports Community (Est.", "date": "", "ddg_snippet": "... has calmly continued where he left off last season, bringing his owners attacking returns on a consistent basis (albeit only during injury time so ...", "subpage_snippet": "", "source": "www.fiso.co.uk", "link": "http://www.fiso.co.uk/", "content": "... has calmly continued where he left off last season, bringing his owners attacking returns on a consistent basis (albeit only during injury time so ..."}
+{"idx": 9, "title": "WO2007149064A1 - Method for tracking using dynamic relational", "date": "", "ddg_snippet": "Instrumental activities of daily living were developed to capture a more complex range of activities, including handling personal finances, meal ...", "subpage_snippet": "", "source": "patents.google.com", "link": "https://patents.google.com/patent/WO2007149064A1/en", "content": "Instrumental activities of daily living were developed to capture a more complex range of activities, including handling personal finances, meal ..."}
diff --git a/data/sampled_jsons/LHRS-Bench_dataset_size_number_of_annotations.jsonl b/data/sampled_jsons/LHRS-Bench_dataset_size_number_of_annotations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..cb3b58bf92b275ae5ec67d3f0c0646f7e14b0531
--- /dev/null
+++ b/data/sampled_jsons/LHRS-Bench_dataset_size_number_of_annotations.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Track changes in your presentation - Microsoft Support", "date": "", "ddg_snippet": "In a collaboration scenario, you could send a review copy of a presentation to other people, collect their changes and comments in that copy, then use the Compare tool in PowerPoint to compare and merge the review copy with your original file. These steps are described in detail below.", "subpage_snippet": "", "source": "support.microsoft.com", "link": "https://support.microsoft.com/en-us/office/track-changes-in-your-presentation-35dad781-50f7-4c4f-9b15-cf418f03c279", "content": "In a collaboration scenario, you could send a review copy of a presentation to other people, collect their changes and comments in that copy, then use the Compare tool in PowerPoint to compare and merge the review copy with your original file. These steps are described in detail below."}
+{"idx": 1, "title": "How to Track Changes in PowerPoint : A Step-By-Step Guide", "date": "", "ddg_snippet": "Feb 16, 2022 · Learning how to track changes can make editing a PowerPoint so much simpler. This article will take you through the essential steps you need to take to track changes within Microsoft PowerPoint .", "subpage_snippet": "", "source": "www.simpleslides.co", "link": "https://www.simpleslides.co/blog/track-changes-in-powerpoint", "content": "Feb 16, 2022 · Learning how to track changes can make editing a PowerPoint so much simpler. This article will take you through the essential steps you need to take to track changes within Microsoft PowerPoint ."}
+{"idx": 2, "title": "How To Track Changes In PowerPoint - Powerpoint Assist", "date": "", "ddg_snippet": "In this comprehensive guide by Oregon-based teacher, Regina Griffin, we will explore the benefits of tracking changes in PowerPoint , who can use this feature, how to enable and track changes , as well as how to review, compare, and share presentations with tracked changes .", "subpage_snippet": "", "source": "pptassist.com", "link": "https://pptassist.com/how-to-track-changes-in-powerpoint/", "content": "In this comprehensive guide by Oregon-based teacher, Regina Griffin, we will explore the benefits of tracking changes in PowerPoint , who can use this feature, how to enable and track changes , as well as how to review, compare, and share presentations with tracked changes ."}
+{"idx": 3, "title": "How to Track Changes in PowerPoint using the Review / Compare...", "date": "", "ddg_snippet": "This blog post steps through how to use the Review feature in PowerPoint to track changes in your PowerPoint presentation. Note that we have used PowerPoint for Microsoft 365, but the steps are similar for other versions of PowerPoint .", "subpage_snippet": "", "source": "pptproductivity.com", "link": "https://pptproductivity.com/blog/how-to-track-changes-in-powerpoint-using-review-compare-feature", "content": "This blog post steps through how to use the Review feature in PowerPoint to track changes in your PowerPoint presentation. Note that we have used PowerPoint for Microsoft 365, but the steps are similar for other versions of PowerPoint ."}
+{"idx": 4, "title": "How to Track Changes in PowerPoint - SlideModel", "date": "", "ddg_snippet": "May 17, 2024 · Just like with Google Slides, you can check who made adjustments to your files in collaborative work. Learn how to track changes in PowerPoint here!", "subpage_snippet": "", "source": "slidemodel.com", "link": "https://slidemodel.com/how-to-track-changes-in-powerpoint/", "content": "May 17, 2024 · Just like with Google Slides, you can check who made adjustments to your files in collaborative work. Learn how to track changes in PowerPoint here!"}
+{"idx": 5, "title": "How to track changes in PowerPoint - Plus", "date": "", "ddg_snippet": "Aug 9, 2024 · When you work on a presentation with others, you want to see the edits they make, right? We’ll show you step-by-step how to track changes in PowerPoint .", "subpage_snippet": "", "source": "plusai.com", "link": "https://plusai.com/blog/how-to-track-changes-in-powerpoint", "content": "Aug 9, 2024 · When you work on a presentation with others, you want to see the edits they make, right? We’ll show you step-by-step how to track changes in PowerPoint ."}
+{"idx": 6, "title": "How to Track Changes in Microsoft PowerPoint - Vegaslide", "date": "", "ddg_snippet": "Jan 3, 2024 · Open the PowerPoint presentation you want to track changes on. On the “Review” tab, click “ Track Changes ”. This turns on tracking and shows the “ Track Changes ” options. Make sure the “Highlight Changes” box is checked off so edits will be clearly marked. Track Changes ribbon in PowerPoint .", "subpage_snippet": "", "source": "vegaslide.com", "link": "https://vegaslide.com/how-to-track-changes-in-microsoft-powerpoint/", "content": "Jan 3, 2024 · Open the PowerPoint presentation you want to track changes on. On the “Review” tab, click “ Track Changes ”. This turns on tracking and shows the “ Track Changes ” options. Make sure the “Highlight Changes” box is checked off so edits will be clearly marked. Track Changes ribbon in PowerPoint ."}
+{"idx": 7, "title": "How to Track Changes in a Microsoft PowerPoint Presentation", "date": "", "ddg_snippet": "In this article, we will discuss how to track changes in a Microsoft PowerPoint presentation, including enabling the Track Changes feature, reviewing changes, accepting or rejecting changes, and sharing the revised presentation with others.", "subpage_snippet": "", "source": "umatechnology.org", "link": "https://umatechnology.org/how-to-track-changes-in-a-microsoft-powerpoint-presentation/", "content": "In this article, we will discuss how to track changes in a Microsoft PowerPoint presentation, including enabling the Track Changes feature, reviewing changes, accepting or rejecting changes, and sharing the revised presentation with others."}
+{"idx": 8, "title": "How to Track Changes in a Microsoft PowerPoint Presentation", "date": "", "ddg_snippet": "Oct 30, 2020 · While Microsoft 365 subscribers can do real-time collaboration on PowerPoint presentations, some people prefer working independently and having that work reviewed and edited at a later date. Here's how you can see and track what changed when the presentation comes back to you.", "subpage_snippet": "", "source": "www.howtogeek.com", "link": "https://www.howtogeek.com/690222/how-to-track-changes-in-microsoft-powerpoint/", "content": "Oct 30, 2020 · While Microsoft 365 subscribers can do real-time collaboration on PowerPoint presentations, some people prefer working independently and having that work reviewed and edited at a later date. Here's how you can see and track what changed when the presentation comes back to you."}
+{"idx": 9, "title": "How to Track Changes in PowerPoint for Better Collaboration", "date": "", "ddg_snippet": "Mar 10, 2025 · Before starting any collaborative editing or review, save a draft of your PowerPoint presentation. Keeping the unedited, original file as your review copy simplifies comparing edits and easily reverts changes . Here’s how: via PowerPoint .", "subpage_snippet": "", "source": "clickup.com", "link": "https://clickup.com/blog/how-to-track-changes-in-powerpoint/", "content": "Mar 10, 2025 · Before starting any collaborative editing or review, save a draft of your PowerPoint presentation. Keeping the unedited, original file as your review copy simplifies comparing edits and easily reverts changes . Here’s how: via PowerPoint ."}
diff --git a/data/sampled_jsons/Large_Language_Model-Brained_GUI_Agents_survey_summary_action_grounding_year_2024.jsonl b/data/sampled_jsons/Large_Language_Model-Brained_GUI_Agents_survey_summary_action_grounding_year_2024.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..69a8e69d75843fe3b6969c859d04f39a9ade3512
--- /dev/null
+++ b/data/sampled_jsons/Large_Language_Model-Brained_GUI_Agents_survey_summary_action_grounding_year_2024.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Large Language Model-Brained GUI Agents: A Survey", "date": "", "ddg_snippet": "27 Nov 2024 — This paper presents a comprehensive survey of LLM-brained GUI agents , exploring their historical evolution, core components, and advanced techniques.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2411.18279v1", "content": "27 Nov 2024 — This paper presents a comprehensive survey of LLM-brained GUI agents , exploring their historical evolution, core components, and advanced techniques."}
+{"idx": 1, "title": "Large Language Model-Brained GUI Agents: A Survey", "date": "", "ddg_snippet": "by C Zhang · Cited by 81 — This paper presents a comprehensive survey of LLM- brained GUI agents , exploring their historical evolution, core components, and advanced techniques.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=xChvYjvXTp", "content": "by C Zhang · Cited by 81 — This paper presents a comprehensive survey of LLM- brained GUI agents , exploring their historical evolution, core components, and advanced techniques."}
+{"idx": 2, "title": "Large Language Model-Brained GUI Agents: A Survey", "date": "", "ddg_snippet": "Intelligent agents that operate within GUI environments , leveraging LLMs as their core inference and cognitive engine to generate, plan, and execute actions in ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2411.18279v11", "content": "Intelligent agents that operate within GUI environments , leveraging LLMs as their core inference and cognitive engine to generate, plan, and execute actions in ..."}
+{"idx": 3, "title": "Large Language Model-Brained GUI Agents: A Survey", "date": "", "ddg_snippet": "27 Nov 2024 — This paper presents a comprehensive survey of LLM- brained GUI agents , exploring their historical evolution, core components, and advanced techniques.", "subpage_snippet": "", "source": "huggingface.co", "link": "https://huggingface.co/papers/2411.18279", "content": "27 Nov 2024 — This paper presents a comprehensive survey of LLM- brained GUI agents , exploring their historical evolution, core components, and advanced techniques."}
+{"idx": 4, "title": "Large Language Model-Brained GUI Agents: A Survey", "date": "", "ddg_snippet": "We address critical research questions such as existing GUI agent frameworks, the collection and utilization of data for training specialized GUI agents , the.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf/0e2a59aacb16c1cf9894b8baf1b419b2e69ad522.pdf", "content": "We address critical research questions such as existing GUI agent frameworks, the collection and utilization of data for training specialized GUI agents , the."}
+{"idx": 5, "title": "GUI Agents: A Survey", "date": "", "ddg_snippet": "by D Nguyen · 2025 · Cited by 43 — This section focuses on various architectural de- signs of a GUI agent , which we categorize into four main types: (1) Perception: designs that ... 17 pages", "subpage_snippet": "", "source": "aclanthology.org", "link": "https://aclanthology.org/2025.findings-acl.1158.pdf", "content": "by D Nguyen · 2025 · Cited by 43 — This section focuses on various architectural de- signs of a GUI agent , which we categorize into four main types: (1) Perception: designs that ... 17 pages"}
+{"idx": 6, "title": "Large Language Model-Brained GUI Agents: A Survey | PDF", "date": "", "ddg_snippet": "This paper surveys the emerging field of Large Language Model (LLM)-brained GUI agents , which automate graphical user interface interactions using natural ...", "subpage_snippet": "", "source": "www.scribd.com", "link": "https://www.scribd.com/document/844194456/10", "content": "This paper surveys the emerging field of Large Language Model (LLM)-brained GUI agents , which automate graphical user interface interactions using natural ..."}
+{"idx": 7, "title": "Large Language Model-Brained GUI Agents: A Survey", "date": "", "ddg_snippet": "27 Nov 2024 — Key takeaway: 'LLM- brained GUI agents revolutionize user experience by enabling complex tasks through simple conversational commands, ...", "subpage_snippet": "", "source": "k8s.consensus.app", "link": "https://k8s.consensus.app/papers/details/1b987bb627245572b0ce96544dc94030/", "content": "27 Nov 2024 — Key takeaway: 'LLM- brained GUI agents revolutionize user experience by enabling complex tasks through simple conversational commands, ..."}
+{"idx": 8, "title": "LLM-Brained GUI Agents", "date": "", "ddg_snippet": "13 Aug 2025 — LLM- brained GUI agents are autonomous systems that combine modular architectures, integrating perception, planning, action , and memory for ...", "subpage_snippet": "", "source": "www.emergentmind.com", "link": "https://www.emergentmind.com/topics/large-language-model-brained-gui-agents", "content": "13 Aug 2025 — LLM- brained GUI agents are autonomous systems that combine modular architectures, integrating perception, planning, action , and memory for ..."}
+{"idx": 9, "title": "Large Language Model-Brained GUI Agents: A Survey ...", "date": "", "ddg_snippet": "To provide a structured understanding of this trend, this paper presents a comprehensive survey of LLM- brained GUI agents , exploring their ...", "subpage_snippet": "", "source": "www.facebook.com", "link": "https://www.facebook.com/groups/DeepNetGroup/posts/2346552319070970/", "content": "To provide a structured understanding of this trend, this paper presents a comprehensive survey of LLM- brained GUI agents , exploring their ..."}
diff --git a/data/sampled_jsons/Lattimore_Szepesvari_partial_monitoring_optimism_abstract.jsonl b/data/sampled_jsons/Lattimore_Szepesvari_partial_monitoring_optimism_abstract.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..a6226a705b854c95595194e4fa6ac45dfa43bee6
--- /dev/null
+++ b/data/sampled_jsons/Lattimore_Szepesvari_partial_monitoring_optimism_abstract.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Exploration by Optimisation in Partial Monitoring", "date": "", "ddg_snippet": "Exploration by Optimisation in Partial MonitoringTor Lattimore , Csaba SzepesváriWe provide a novel algorithm for adversarial k-action d-outcome partial ...", "subpage_snippet": "", "source": "proceedings.mlr.press", "link": "https://proceedings.mlr.press/v125/lattimore20a.html", "content": "Exploration by Optimisation in Partial MonitoringTor Lattimore , Csaba SzepesváriWe provide a novel algorithm for adversarial k-action d-outcome partial ..."}
+{"idx": 1, "title": "[1907.05772v2] Exploration by Optimisation in Partial Monitoring", "date": "", "ddg_snippet": "View a PDF of the paper titled Exploration by Optimisation in Partial Monitoring , by Tor Lattimore and Csaba Szepesvari", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/1907.05772v2", "content": "View a PDF of the paper titled Exploration by Optimisation in Partial Monitoring , by Tor Lattimore and Csaba Szepesvari"}
+{"idx": 2, "title": "PDF Linear Partial Monitoring for Sequential Decision Making Algorithms ...", "date": "", "ddg_snippet": "Abstract Partial monitoring is an expressive framework for sequential decision-making with an abun-dance of applications, including graph-structured and dueling bandits, dynamic pricing and transductive feedback models. We survey and extend recent results on the linear formu-lation of partial monitoring that naturally generalizes the standard linear bandit setting. The main result is that a ...", "subpage_snippet": "", "source": "jmlr.org", "link": "https://jmlr.org/papers/volume24/22-1248/22-1248.pdf", "content": "Abstract Partial monitoring is an expressive framework for sequential decision-making with an abun-dance of applications, including graph-structured and dueling bandits, dynamic pricing and transductive feedback models. We survey and extend recent results on the linear formu-lation of partial monitoring that naturally generalizes the standard linear bandit setting. The main result is that a ..."}
+{"idx": 3, "title": "COLT 2020: Exploration by Optimisation in Partial Monitoring", "date": "", "ddg_snippet": "Abstract : We provide a novel algorithm for adversarial k-action d-outcome partial monitoring that is adaptive, intuitive and efficient. The highlight is that for the non-degenerate locally observable games, the n-round minimax regret is bounded by 2mk^ (3/2)sqrt (3n log (k)), where m is the number of signals. This matches the best known information-theoretic upper bound derived via Bayesian ...", "subpage_snippet": "", "source": "www.learningtheory.org", "link": "https://www.learningtheory.org/colt2020/virtual/papers/paper_66.html", "content": "Abstract : We provide a novel algorithm for adversarial k-action d-outcome partial monitoring that is adaptive, intuitive and efficient. The highlight is that for the non-degenerate locally observable games, the n-round minimax regret is bounded by 2mk^ (3/2)sqrt (3n log (k)), where m is the number of signals. This matches the best known information-theoretic upper bound derived via Bayesian ..."}
+{"idx": 4, "title": "PDF Partial monitoring - tor-lattimore.com", "date": "", "ddg_snippet": "Partial monitoring is a generalisation of the bandit framework that decouples the loss and the obser-vations. The framework is sufficiently rich to model bandits, linear bandits, full information games, dynamic pricing, bandits with graph feedback and many problems between and beyond these exam-ples.", "subpage_snippet": "", "source": "tor-lattimore.com", "link": "https://tor-lattimore.com/downloads/papers/2019-pm-simple.pdf", "content": "Partial monitoring is a generalisation of the bandit framework that decouples the loss and the obser-vations. The framework is sufficiently rich to model bandits, linear bandits, full information games, dynamic pricing, bandits with graph feedback and many problems between and beyond these exam-ples."}
+{"idx": 5, "title": "Exploration by Optimisation in Partial Monitoring | Request PDF", "date": "", "ddg_snippet": "Recently, this condition has been shown by (Bartok, Pal, and Szepesvari , 2011) to imply the O (\\sqrt {T}) rate for partial monitoring games against an i.i.d. opponent, and the authors conjectured ...", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/334457375_Exploration_by_Optimisation_in_Partial_Monitoring", "content": "Recently, this condition has been shown by (Bartok, Pal, and Szepesvari , 2011) to imply the O (\\sqrt {T}) rate for partial monitoring games against an i.i.d. opponent, and the authors conjectured ..."}
+{"idx": 6, "title": "[1610.04491] The End of Optimism? An Asymptotic Analysis of Finite ...", "date": "", "ddg_snippet": "View a PDF of the paper titled The End of Optimism ? An Asymptotic Analysis of Finite-Armed Linear Bandits, by Tor Lattimore and Csaba Szepesvari", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/1610.04491", "content": "View a PDF of the paper titled The End of Optimism ? An Asymptotic Analysis of Finite-Armed Linear Bandits, by Tor Lattimore and Csaba Szepesvari"}
+{"idx": 7, "title": "PDF Partial monitoring - proceedings.mlr.press", "date": "", "ddg_snippet": "Partial monitoring is a generalisation of the bandit framework that decouples the loss and the obser-vations. The framework is sufficiently rich to model bandits, linear bandits, full information games, dynamic pricing, bandits with graph feedback and many problems between and beyond these ex-amples.", "subpage_snippet": "", "source": "proceedings.mlr.press", "link": "https://proceedings.mlr.press/v125/lattimore20a/lattimore20a.pdf", "content": "Partial monitoring is a generalisation of the bandit framework that decouples the loss and the obser-vations. The framework is sufficiently rich to model bandits, linear bandits, full information games, dynamic pricing, bandits with graph feedback and many problems between and beyond these ex-amples."}
+{"idx": 8, "title": "Linear Partial Monitoring for Sequential Decision Making: Algorithms ...", "date": "", "ddg_snippet": "Abstract Partial monitoring is an expressive framework for sequential decision-making with an abundance of applications, including graph-structured and dueling bandits, dynamic pricing and transductive feedback models.", "subpage_snippet": "", "source": "jmlr.org", "link": "https://jmlr.org/beta/papers/v24/22-1248.html", "content": "Abstract Partial monitoring is an expressive framework for sequential decision-making with an abundance of applications, including graph-structured and dueling bandits, dynamic pricing and transductive feedback models."}
+{"idx": 9, "title": "Linear partial monitoring for sequential decision making:", "date": "", "ddg_snippet": "Abstract Partial monitoring is an expressive framework for sequential decision-making with an abundance of applications, including graph-structured and dueling bandits, dynamic pricing and transductive feedback models.", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/10.5555/3648699.3649045", "content": "Abstract Partial monitoring is an expressive framework for sequential decision-making with an abundance of applications, including graph-structured and dueling bandits, dynamic pricing and transductive feedback models."}
diff --git a/data/sampled_jsons/Leveraging_Per-Instance_Privacy_for_Machine_Unlearning_arXiv_full_paper.jsonl b/data/sampled_jsons/Leveraging_Per-Instance_Privacy_for_Machine_Unlearning_arXiv_full_paper.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..ba4613a916d43a4a3d936743edffcd2498b4bf06
--- /dev/null
+++ b/data/sampled_jsons/Leveraging_Per-Instance_Privacy_for_Machine_Unlearning_arXiv_full_paper.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Leveraging Per-Instance Privacy for Machine Unlearning", "date": "", "ddg_snippet": "by NM Sepahvand · 2025 — This paper presents a per-instance approach to quantifying unlearning difficulty by using per-instance privacy losses, which bounds divergence ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2505.18786", "content": "by NM Sepahvand · 2025 — This paper presents a per-instance approach to quantifying unlearning difficulty by using per-instance privacy losses, which bounds divergence ..."}
+{"idx": 1, "title": "Leveraging Per-Instance Privacy for Machine Unlearning", "date": "", "ddg_snippet": "24 May 2025 — This paper presents work whose goal is to advance the field of machine unlearning , which is specifically oriented to improve the ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2505.18786v1", "content": "24 May 2025 — This paper presents work whose goal is to advance the field of machine unlearning , which is specifically oriented to improve the ..."}
+{"idx": 2, "title": "Leveraging Per-Instance Privacy for Machine Unlearning", "date": "", "ddg_snippet": "This paper presents a theoretically grounded and empirically validated framework for quantifying per - instance unlearning difficulty via privacy loss, with ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=0A4Y9qRnu9&noteId=Zd6KsMzKb8", "content": "This paper presents a theoretically grounded and empirically validated framework for quantifying per - instance unlearning difficulty via privacy loss, with ..."}
+{"idx": 3, "title": "Leveraging Per-Instance Privacy for Machine Unlearning", "date": "", "ddg_snippet": "This paper introduces a per-instance approach to quantify machine unlearning difficulty using per-instance privacy losses, which reliably predict unlearning ...", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/poster/46697", "content": "This paper introduces a per-instance approach to quantify machine unlearning difficulty using per-instance privacy losses, which reliably predict unlearning ..."}
+{"idx": 4, "title": "Leveraging Machine-Unlearning on Pretrained Language ...", "date": "", "ddg_snippet": "30 Aug 2024 — In this paper , we combine the notion of poisoning a pre-trained LLM and causing privacy leakage of the fine-tuned model. More specifically, we ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2408.17354v1", "content": "30 Aug 2024 — In this paper , we combine the notion of poisoning a pre-trained LLM and causing privacy leakage of the fine-tuned model. More specifically, we ..."}
+{"idx": 5, "title": "Leveraging Per-Instance Privacy for Machine Unlearning", "date": "", "ddg_snippet": "This research paper explores how to effectively \"unlearn\" certain information from machine learning models , which is important when data needs to be deleted ...", "subpage_snippet": "", "source": "bytez.com", "link": "https://bytez.com/docs/icml/46697/references?_c=eyJ2IjoxLCJyZWxhdGVkIjpbImNvZGUiLCJyZWZlcmVuY2VzIiwiY29uZmVyZW5jZSJdfQ==", "content": "This research paper explores how to effectively \"unlearn\" certain information from machine learning models , which is important when data needs to be deleted ..."}
+{"idx": 6, "title": "Leveraging Machine-Unlearning on Pretrained Language ...", "date": "", "ddg_snippet": "by MRU Rashid · Cited by 11 — In this paper , we combine the notion of poisoning a pre-trained LLM and causing privacy leakage of the fine-tuned model. More specifically, we.", "subpage_snippet": "", "source": "www.merl.com", "link": "https://www.merl.com/publications/docs/TR2024-168.pdf", "content": "by MRU Rashid · Cited by 11 — In this paper , we combine the notion of poisoning a pre-trained LLM and causing privacy leakage of the fine-tuned model. More specifically, we."}
+{"idx": 7, "title": "Machine Unlearning: Challenges in Data Quality and Access", "date": "", "ddg_snippet": "by M Xu · Cited by 5 — Machine unlearning aims to remove specific knowledge from a well-trained machine learning model. This topic has gained significant attention.", "subpage_snippet": "", "source": "www.ijcai.org", "link": "https://www.ijcai.org/proceedings/2024/0987.pdf", "content": "by M Xu · Cited by 5 — Machine unlearning aims to remove specific knowledge from a well-trained machine learning model. This topic has gained significant attention."}
+{"idx": 8, "title": "Berivan Isik", "date": "", "ddg_snippet": "We introduce a per - instance view of unlearning : per - instance privacy losses to measure how difficult it is to unlearn each point.", "subpage_snippet": "", "source": "x.com", "link": "https://x.com/BerivanISIK/status/1928475964620857800", "content": "We introduce a per - instance view of unlearning : per - instance privacy losses to measure how difficult it is to unlearn each point."}
+{"idx": 9, "title": "Computer Science May 2025", "date": "", "ddg_snippet": "Full text. Search. arXiv logo · Cornell ... Title: Leveraging Per-Instance Privacy for Machine Unlearning ... Comments: This paper was presented at the 16th ...", "subpage_snippet": "", "source": "www.arxiv.org", "link": "https://www.arxiv.org/list/cs/2025-05?skip=9450&show=500", "content": "Full text. Search. arXiv logo · Cornell ... Title: Leveraging Per-Instance Privacy for Machine Unlearning ... Comments: This paper was presented at the 16th ..."}
diff --git a/data/sampled_jsons/MSN_metric_image_segmentation_entity_inpainting.jsonl b/data/sampled_jsons/MSN_metric_image_segmentation_entity_inpainting.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..b5bb6aa746b8c472aae783329df72b55efd4e1ef
--- /dev/null
+++ b/data/sampled_jsons/MSN_metric_image_segmentation_entity_inpainting.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "EntityErasure: Erasing Entity Cleanly via Amodal Entity ...", "date": "", "ddg_snippet": "Next, the amodal entity segmentation , masked image latent ci, inpainting mask, and the noisy latent zt will be fed to Amodal Entity Completion Model (AECM) to predict the final amodal entity completion result.", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content/CVPR2025/papers/Zhu_EntityErasure_Erasing_Entity_Cleanly_via_Amodal_Entity_Segmentation_and_Completion_CVPR_2025_paper.pdf", "content": "Next, the amodal entity segmentation , masked image latent ci, inpainting mask, and the noisy latent zt will be fed to Amodal Entity Completion Model (AECM) to predict the final amodal entity completion result."}
+{"idx": 1, "title": "CVPR Poster EntityErasure: Erasing Entity Cleanly via Amodal ...", "date": "", "ddg_snippet": "MSN reflects the quantity of sundries generated by the model, while MARS indicates the ratio of the sundries area relative to the inpainting mask area. A higher ...", "subpage_snippet": "", "source": "cvpr.thecvf.com", "link": "https://cvpr.thecvf.com/virtual/2025/poster/34016", "content": "MSN reflects the quantity of sundries generated by the model, while MARS indicates the ratio of the sundries area relative to the inpainting mask area. A higher ..."}
+{"idx": 2, "title": "EntityErasure: Erasing Entity Cleanly via Amodal ...", "date": "", "ddg_snippet": "This paper presents EntityErasure , a novel diffusion-based inpainting method that can effectively erase entities without inducing unwanted sundries.", "subpage_snippet": "", "source": "chatpaper.com", "link": "https://chatpaper.com/paper/153180", "content": "This paper presents EntityErasure , a novel diffusion-based inpainting method that can effectively erase entities without inducing unwanted sundries."}
+{"idx": 3, "title": "satellite-image-deep-learning/techniques", "date": "", "ddg_snippet": "This repository provides an exhaustive overview of deep learning techniques specifically tailored for satellite and aerial image processing.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/satellite-image-deep-learning/techniques", "content": "This repository provides an exhaustive overview of deep learning techniques specifically tailored for satellite and aerial image processing."}
+{"idx": 4, "title": "Comprehensive analyses of image forgery detection ...", "date": "", "ddg_snippet": "by P Sharma · 2023 · Cited by 80 — This paper overviews the evaluation of various image tamper detection methods. A brief discussion of image datasets and a comparative study of ...", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/10.1007/s11042-022-13808-w", "content": "by P Sharma · 2023 · Cited by 80 — This paper overviews the evaluation of various image tamper detection methods. A brief discussion of image datasets and a comparative study of ..."}
+{"idx": 5, "title": "EASY Inpainting in ComfyUI with SAM ( segment Anything) - YouTube", "date": "", "ddg_snippet": "Ready to take your image editing skills to the next level? Join me in this journey as we uncover the most mind-blowing inpainting techniques you won't believ...", "subpage_snippet": "", "source": "www.youtube.com", "link": "https://www.youtube.com/watch?v=SMOM1bIY5yA", "content": "Ready to take your image editing skills to the next level? Join me in this journey as we uncover the most mind-blowing inpainting techniques you won't believ..."}
+{"idx": 6, "title": "Improving Performance of Image Segmentation with... | Medium", "date": "", "ddg_snippet": "For instance, the Dice coefficient is a popular image segmentation metric used in performance estimation. Nevertheless, after careful manual examination of the prediction masks, small “islands” of mispredicted pixels become visible.", "subpage_snippet": "", "source": "medium.com", "link": "https://medium.com/@ihor.shylo/improving-performance-of-image-segmentation-with-conditional-random-fields-crf-8b93f7db396c", "content": "For instance, the Dice coefficient is a popular image segmentation metric used in performance estimation. Nevertheless, after careful manual examination of the prediction masks, small “islands” of mispredicted pixels become visible."}
+{"idx": 7, "title": "Master Inpainting on Flux & SegmentAnything2 with... - Endangered Ai", "date": "", "ddg_snippet": "Inpainting with the Flux model and enhancing your workflow with Facebook’s Segment Anything tool offers a powerful combination for AI-driven image editing.", "subpage_snippet": "", "source": "endangeredai.com", "link": "https://endangeredai.com/master-inpainting-on-flux-segmentanything2-with-comfyui-what-you-need-to-know-for-outstanding-results/", "content": "Inpainting with the Flux model and enhancing your workflow with Facebook’s Segment Anything tool offers a powerful combination for AI-driven image editing."}
+{"idx": 8, "title": "What is the Intersection over Union (IoU) metric used for in image ...", "date": "", "ddg_snippet": "Intersection over Union (IoU) is a crucial evaluation metric in tasks like object detection and image segmentation . It measures the overlap between predicted bo.", "subpage_snippet": "", "source": "askai.glarity.app", "link": "https://askai.glarity.app/search/What-is-the-Intersection-over-Union--IoU--metric-used-for-in-image-segmentation-validation-functions", "content": "Intersection over Union (IoU) is a crucial evaluation metric in tasks like object detection and image segmentation . It measures the overlap between predicted bo."}
+{"idx": 9, "title": "Image Inpainting — MCQs | Digital Image Processing...", "date": "", "ddg_snippet": "(A) Image segmentation (B) Object detection (C) Restoration of missing or damaged regions (D) Enhancement of image resolution. 3. Which mathematical tool is frequently used in PDE-based image inpainting ?", "subpage_snippet": "", "source": "t4tutorials.com", "link": "https://t4tutorials.com/image-inpainting-mcqs-digital-image-processing/", "content": "(A) Image segmentation (B) Object detection (C) Restoration of missing or damaged regions (D) Enhancement of image resolution. 3. Which mathematical tool is frequently used in PDE-based image inpainting ?"}
diff --git a/data/sampled_jsons/Machine_Learning_meets_Algebraic_Combinatorics_spurious_correlation_0.98.jsonl b/data/sampled_jsons/Machine_Learning_meets_Algebraic_Combinatorics_spurious_correlation_0.98.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..96520797946438153b22ba791a850b1bb856ecd0
--- /dev/null
+++ b/data/sampled_jsons/Machine_Learning_meets_Algebraic_Combinatorics_spurious_correlation_0.98.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "SpurBreast: A Curated Dataset for Investigating Spurious ...", "date": "", "ddg_snippet": "It includes two experimental datasets with specific biases to evaluate model robustness. The first dataset intro-duces a spurious correlation with MRI magnetic field strength and the second dataset introduces spurious correlations based on vertical alignment.", "subpage_snippet": "", "source": "link.springer.com", "link": "https://link.springer.com/content/pdf/10.1007/978-3-032-05325-1_53.pdf?pdf=inline+link", "content": "It includes two experimental datasets with specific biases to evaluate model robustness. The first dataset intro-duces a spurious correlation with MRI magnetic field strength and the second dataset introduces spurious correlations based on vertical alignment."}
+{"idx": 1, "title": "ACE and Diverse Generalization via Selective Disagreement", "date": "", "ddg_snippet": "Using a self-training approach that encourages confident and selective disagreement, our method ACE matches or outperforms existing methods on a suite of complete- spurious correlation benchmarks, while remaining robust to incomplete spurious correlations.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2509.07955", "content": "Using a self-training approach that encourages confident and selective disagreement, our method ACE matches or outperforms existing methods on a suite of complete- spurious correlation benchmarks, while remaining robust to incomplete spurious correlations."}
+{"idx": 2, "title": "Daily Papers - Hugging Face", "date": "", "ddg_snippet": "With the help of in-context learning (ICL), large language models (LLMs) have achieved impressive performance across various tasks. However, the function of descriptive instructions during ICL remains under-explored. In this work, we propose an ensemble prompt framework to describe the selection criteria of multiple in-context examples, and preliminary experiments on machine translation (MT ...", "subpage_snippet": "", "source": "huggingface.co", "link": "https://huggingface.co/papers?q=automated+ensembles", "content": "With the help of in-context learning (ICL), large language models (LLMs) have achieved impressive performance across various tasks. However, the function of descriptive instructions during ICL remains under-explored. In this work, we propose an ensemble prompt framework to describe the selection criteria of multiple in-context examples, and preliminary experiments on machine translation (MT ..."}
+{"idx": 3, "title": "Hot Topics and Popular Papers in Evolutionary Psychology ...", "date": "", "ddg_snippet": "The only remaining limitation is that we cannot rule out a spurious correlation based on unknown and untested “third” variables. In addition, the moderately strong effect size of the reference-citation relationship suggests that it may warrant further study.", "subpage_snippet": "", "source": "journals.sagepub.com", "link": "https://journals.sagepub.com/doi/10.1177/147470490900700301?int.sj-full-text.similar-articles.9", "content": "The only remaining limitation is that we cannot rule out a spurious correlation based on unknown and untested “third” variables. In addition, the moderately strong effect size of the reference-citation relationship suggests that it may warrant further study."}
+{"idx": 4, "title": "FrontierMath: A Benchmark for Evaluating Advanced Mathematical", "date": "", "ddg_snippet": "... algebraic geometry, number theory, point set and algebraic topology, combinatorics , category theory, representation theory, partial differential ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2411.04872v5", "content": "... algebraic geometry, number theory, point set and algebraic topology, combinatorics , category theory, representation theory, partial differential ..."}
+{"idx": 5, "title": "Does Math Reasoning Improve General LLM Capabilities?", "date": "", "ddg_snippet": "Mathematics is often considered the foundational language of science, and enabling machines to reason precisely over math is central to the long-term ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2507.00432v1", "content": "Mathematics is often considered the foundational language of science, and enabling machines to reason precisely over math is central to the long-term ..."}
+{"idx": 6, "title": "Artificial Intelligence Policy Uncertainty and Corporate ...", "date": "", "ddg_snippet": "6 days ago · While the economic consequences of general policy uncertainty are well-documented, little is known about how uncertainty surrounding the governance of…", "subpage_snippet": "", "source": "www.sciencedirect.com", "link": "https://www.sciencedirect.com/science/article/pii/S1059056025007932", "content": "6 days ago · While the economic consequences of general policy uncertainty are well-documented, little is known about how uncertainty surrounding the governance of…"}
+{"idx": 7, "title": "Popularity of the first name Pauline correlates with Violent ...", "date": "", "ddg_snippet": "Data details Popularity of the first name PaulineDetailed data title: Babies of all sexes born in the US named Pauline Source: US Social Security Administration See what else correlates with Popularity of the first name PaulineViolent crime ratesDetailed data title: The violent crime rate per 100,000 residents in United States Source: FBI Criminal Justice Information Services See what else ...", "subpage_snippet": "", "source": "www.tylervigen.com", "link": "http://www.tylervigen.com/spurious/correlation/24489_popularity-of-the-first-name-pauline_correlates-with_violent-crime-rates", "content": "Data details Popularity of the first name PaulineDetailed data title: Babies of all sexes born in the US named Pauline Source: US Social Security Administration See what else correlates with Popularity of the first name PaulineViolent crime ratesDetailed data title: The violent crime rate per 100,000 residents in United States Source: FBI Criminal Justice Information Services See what else ..."}
+{"idx": 8, "title": "Multimodal trajectory prediction for intelligent connected ...", "date": "", "ddg_snippet": "The safe and efficient operation of smart Intelligent vehicles relies heavily on accurate trajectory prediction techniques. Existing methods improve prediction accuracy by introducing scene ...", "subpage_snippet": "", "source": "preview-www.nature.com", "link": "https://preview-www.nature.com/articles/s41598-025-91818-y.pdf", "content": "The safe and efficient operation of smart Intelligent vehicles relies heavily on accurate trajectory prediction techniques. Existing methods improve prediction accuracy by introducing scene ..."}
+{"idx": 9, "title": "Hogg's Research", "date": "", "ddg_snippet": "After the meeting , I implemented some of that, but problems like this have a pathology: If you carefully remove stars with high residuals at some ...", "subpage_snippet": "", "source": "hoggresearch.blogspot.com", "link": "https://hoggresearch.blogspot.com/", "content": "After the meeting , I implemented some of that, but problems like this have a pathology: If you carefully remove stars with high residuals at some ..."}
diff --git a/data/sampled_jsons/Medusa_Simple_LLM_Inference_Acceleration_Framework_with_Multiple_Decoding_Heads_Cai_et_al.jsonl b/data/sampled_jsons/Medusa_Simple_LLM_Inference_Acceleration_Framework_with_Multiple_Decoding_Heads_Cai_et_al.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..8b514507077b0aee3e88a6b7171309afe55a0cba
--- /dev/null
+++ b/data/sampled_jsons/Medusa_Simple_LLM_Inference_Acceleration_Framework_with_Multiple_Decoding_Heads_Cai_et_al.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Medusa: Simple LLM Inference Acceleration Framework with ... MEDUSA | Proceedings of the 41st International Conference on ... Medusa: Simple Framework for Accelerating LLM Generation with ... [Paper Reading] Medusa: Simple LLM Inference Acceleration ... Medusa: Multiple Decoding Heads for Faster LLM Inference Medusa: Simple LLM Inference Acceleration Framework with ... MEDUSA | Proceedings of the 41st International Conference on Machine Medusa : Simple LLM Inference Acceleration Framework with Multiple GitHub - FasterDecoding/ Medusa : Medusa : Simple Framework for Medusa : Simple LLM Inference Acceleration Framework with Multiple MEDUSA | Proceedings of the 41st International Conference on Machine [Paper Reading] Medusa : Simple LLM Inference Acceleration Framework Abstract MEDUSA: Simple LLM Inference Acceleration Framework ...", "date": "", "ddg_snippet": "Jan 19, 2024 · In this paper, we present Medusa , an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. In this paper, we present MEDUSA , an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Medusa is a simple framework that democratizes the acceleration techniques for LLM generation with multiple decoding heads . Medusa -1 on Vicuna-7b. We aim to tackle the three pain points of popular acceleration techniques like speculative decoding : •Requirement of a good draft model. •System complexity. •Inefficiency when using sampling-based generation. See full list on github.com See full list on github.com medusa /model/medusa_model.py is the key file for Medusa . It contains the MedusaModel class, which is a wrapper of the original model and the new heads . This class also has an implementation of a streaming generation method. If you want to dive into the details of Medusa , this is the place to start. We also provide some illustrative notebooks in notebooks/ to help you understand the codebase. See full list on github.com We are super excited to see that Medusa has been adopted by many open-source projects. Here is an (incomplete) list: •TensorRT- LLM •TGI •RTP- LLM See full list on github.com We welcome community contributions to Medusa . If you have an idea for how to improve it, please open an issue to discuss it with us. When submitting a pull request, please ensure that your changes are well-tested. Please split each major change into a separate pull request. We also have a Roadmap summarizing our future plans for Medusa . Don't hesit... See full list on github.com This codebase is influenced by remarkable projects from the LLM community, including FastChat, TinyChat, vllm, axolotl. This project is supported by Together AI, MyShell AI, Chai AI. See full list on github.com Jul 1, 2025 · Today, I’m introducing a classic multi - head decoding architecture called Medusa . Medusa , inspired by the mythological Greek figure also known as the “snake-haired woman,” has each decoding head metaphorically representing a snake. The architecture mirrors this imagery with multiple decoding heads . Medusa seeks to improve model inference speed by adding additional decoding heads to existing LLMs . These additional heads enable multi-token prediction and can be trained with the backbone LLM frozen, or in conjunction with the original LLM. Read the paper: https://arxiv.org/abs/2401.10774. Specifically, we introduce MEDUSA , a method that enhances LLM inference by integrating additional decoding heads to concurrently predict multiple tokens. These heads are fine-tuned in a parameter-eficient manner and can be added to any existing model. Can Medusa augment LLM inference by adding additional decoding heads? In this paper, we present MEDUSA, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, MEDUSA constructs multiple candidate continuations and verifies them simultaneously in each decoding step. How does Medusa augment LLM inference? In this paper, we present Medusa, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, Medusa constructs multiple candidate continuations and verifies them simultaneously in each decoding step. Does Medusa speed up speculative decoding? The new results show a 2.2-3.6x speedup over the original model on a range of LLMs. Medusa is a simple framework that democratizes the acceleration techniques for LLM generation with multiple decoding heads . Medusa -1 on Vicuna-7b. We aim to tackle the three pain points of popular acceleration techniques like speculative decoding : How does Medusa reduce the number of decoding steps required? By leveraging parallel processing , Medusa substantially reduces the number of decoding steps required. We present two levels of fine-tuning procedures for Medusa to meet the needs of different use cases: Medusa-1: Medusa is directly fine-tuned on top of a frozen backbone LLM, enabling lossless inference acceleration. What's the difference between Medusa 2 & LLM? MEDUSA -2: MEDUSA is fine-tuned together with the backbone LLM , enabling better prediction accuracy of MEDUSA heads and higher speedup but needing a special training recipe that preserves the model's capabilities. What are the components of Medusa? The two most crucial components in Medusa are the multiple decoding heads and the tree-based attention mechanism . The decoding head structure is shown below: In practice, retraining the entire decoding layer may introduce too many parameters (head_num * last_hidden_size * vocab_size). MEDUSA -1: MEDUSA is directly fine-tuned on top of a frozen backbone LLM , enabling lossless inference acceleration .", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2401.10774", "content": "Jan 19, 2024 · In this paper, we present Medusa , an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. In this paper, we present MEDUSA , an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Medusa is a simple framework that democratizes the acceleration techniques for LLM generation with multiple decoding heads . Medusa -1 on Vicuna-7b. We aim to tackle the three pain points of popular acceleration techniques like speculative decoding : •Requirement of a good draft model. •System complexity. •Inefficiency when using sampling-based generation. See full list on github.com See full list on github.com medusa /model/medusa_model.py is the key file for Medusa . It contains the MedusaModel class, which is a wrapper of the original model and the new heads . This class also has an implementation of a streaming generation method. If you want to dive into the details of Medusa , this is the place to start. We also provide some illustrative notebooks in notebooks/ to help you understand the codebase. See full list on github.com We are super excited to see that Medusa has been adopted by many open-source projects. Here is an (incomplete) list: •TensorRT- LLM •TGI •RTP- LLM See full list on github.com We welcome community contributions to Medusa . If you have an idea for how to improve it, please open an issue to discuss it with us. When submitting a pull request, please ensure that your changes are well-tested. Please split each major change into a separate pull request. We also have a Roadmap summarizing our future plans for Medusa . Don't hesit... See full list on github.com This codebase is influenced by remarkable projects from the LLM community, including FastChat, TinyChat, vllm, axolotl. This project is supported by Together AI, MyShell AI, Chai AI. See full list on github.com Jul 1, 2025 · Today, I’m introducing a classic multi - head decoding architecture called Medusa . Medusa , inspired by the mythological Greek figure also known as the “snake-haired woman,” has each decoding head metaphorically representing a snake. The architecture mirrors this imagery with multiple decoding heads . Medusa seeks to improve model inference speed by adding additional decoding heads to existing LLMs . These additional heads enable multi-token prediction and can be trained with the backbone LLM frozen, or in conjunction with the original LLM. Read the paper: https://arxiv.org/abs/2401.10774. Specifically, we introduce MEDUSA , a method that enhances LLM inference by integrating additional decoding heads to concurrently predict multiple tokens. These heads are fine-tuned in a parameter-eficient manner and can be added to any existing model. Can Medusa augment LLM inference by adding additional decoding heads? In this paper, we present MEDUSA, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, MEDUSA constructs multiple candidate continuations and verifies them simultaneously in each decoding step. How does Medusa augment LLM inference? In this paper, we present Medusa, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, Medusa constructs multiple candidate continuations and verifies them simultaneously in each decoding step. Does Medusa speed up speculative decoding? The new results show a 2.2-3.6x speedup over the original model on a range of LLMs. Medusa is a simple framework that democratizes the acceleration techniques for LLM generation with multiple decoding heads . Medusa -1 on Vicuna-7b. We aim to tackle the three pain points of popular acceleration techniques like speculative decoding : How does Medusa reduce the number of decoding steps required? By leveraging parallel processing , Medusa substantially reduces the number of decoding steps required. We present two levels of fine-tuning procedures for Medusa to meet the needs of different use cases: Medusa-1: Medusa is directly fine-tuned on top of a frozen backbone LLM, enabling lossless inference acceleration. What's the difference between Medusa 2 & LLM? MEDUSA -2: MEDUSA is fine-tuned together with the backbone LLM , enabling better prediction accuracy of MEDUSA heads and higher speedup but needing a special training recipe that preserves the model's capabilities. What are the components of Medusa? The two most crucial components in Medusa are the multiple decoding heads and the tree-based attention mechanism . The decoding head structure is shown below: In practice, retraining the entire decoding layer may introduce too many parameters (head_num * last_hidden_size * vocab_size). MEDUSA -1: MEDUSA is directly fine-tuned on top of a frozen backbone LLM , enabling lossless inference acceleration ."}
+{"idx": 1, "title": "MEDUSA: Simple LLM inference acceleration framework ...", "date": "", "ddg_snippet": "by T Cai · 2024 · Cited by 348 — In this paper, we present MEDUSA , an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in ...", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/10.5555/3692070.3692273", "content": "by T Cai · 2024 · Cited by 348 — In this paper, we present MEDUSA , an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in ..."}
+{"idx": 2, "title": "Medusa: Simple Framework for Accelerating LLM Generation with ... [Paper Reading] Medusa: Simple LLM Inference Acceleration ... Medusa: Multiple Decoding Heads for Faster LLM Inference Medusa: Simple LLM Inference Acceleration Framework with ... MEDUSA | Proceedings of the 41st International Conference on Machine Medusa : Simple LLM Inference Acceleration Framework with Multiple GitHub - FasterDecoding/ Medusa : Medusa : Simple Framework for Medusa : Simple LLM Inference Acceleration Framework with Multiple MEDUSA | Proceedings of the 41st International Conference on Machine [Paper Reading] Medusa : Simple LLM Inference Acceleration Framework Abstract MEDUSA: Simple LLM Inference Acceleration Framework ...", "date": "", "ddg_snippet": "Medusa is a simple framework that democratizes the acceleration techniques for LLM generation with multiple decoding heads . Medusa -1 on Vicuna-7b. We aim to tackle the three pain points of popular acceleration techniques like speculative decoding : •Requirement of a good draft model. •System complexity. •Inefficiency when using sampling-based generation. See full list on github.com See full list on github.com medusa /model/medusa_model.py is the key file for Medusa . It contains the MedusaModel class, which is a wrapper of the original model and the new heads . This class also has an implementation of a streaming generation method. If you want to dive into the details of Medusa , this is the place to start. We also provide some illustrative notebooks in notebooks/ to help you understand the codebase. See full list on github.com We are super excited to see that Medusa has been adopted by many open-source projects. Here is an (incomplete) list: •TensorRT- LLM •TGI •RTP- LLM See full list on github.com We welcome community contributions to Medusa . If you have an idea for how to improve it, please open an issue to discuss it with us. When submitting a pull request, please ensure that your changes are well-tested. Please split each major change into a separate pull request. We also have a Roadmap summarizing our future plans for Medusa . Don't hesit... See full list on github.com This codebase is influenced by remarkable projects from the LLM community, including FastChat, TinyChat, vllm, axolotl. This project is supported by Together AI, MyShell AI, Chai AI. See full list on github.com Jul 1, 2025 · Today, I’m introducing a classic multi - head decoding architecture called Medusa . Medusa , inspired by the mythological Greek figure also known as the “snake-haired woman,” has each decoding head metaphorically representing a snake. The architecture mirrors this imagery with multiple decoding heads . Medusa seeks to improve model inference speed by adding additional decoding heads to existing LLMs . These additional heads enable multi-token prediction and can be trained with the backbone LLM frozen, or in conjunction with the original LLM. Read the paper: https://arxiv.org/abs/2401.10774. Specifically, we introduce MEDUSA , a method that enhances LLM inference by integrating additional decoding heads to concurrently predict multiple tokens. These heads are fine-tuned in a parameter-eficient manner and can be added to any existing model. Can Medusa augment LLM inference by adding additional decoding heads? In this paper, we present MEDUSA, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, MEDUSA constructs multiple candidate continuations and verifies them simultaneously in each decoding step. How does Medusa augment LLM inference? In this paper, we present Medusa, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, Medusa constructs multiple candidate continuations and verifies them simultaneously in each decoding step. Does Medusa speed up speculative decoding? The new results show a 2.2-3.6x speedup over the original model on a range of LLMs. Medusa is a simple framework that democratizes the acceleration techniques for LLM generation with multiple decoding heads . Medusa -1 on Vicuna-7b. We aim to tackle the three pain points of popular acceleration techniques like speculative decoding : How does Medusa reduce the number of decoding steps required? By leveraging parallel processing , Medusa substantially reduces the number of decoding steps required. We present two levels of fine-tuning procedures for Medusa to meet the needs of different use cases: Medusa-1: Medusa is directly fine-tuned on top of a frozen backbone LLM, enabling lossless inference acceleration. What's the difference between Medusa 2 & LLM? MEDUSA -2: MEDUSA is fine-tuned together with the backbone LLM , enabling better prediction accuracy of MEDUSA heads and higher speedup but needing a special training recipe that preserves the model's capabilities. What are the components of Medusa? The two most crucial components in Medusa are the multiple decoding heads and the tree-based attention mechanism . The decoding head structure is shown below: In practice, retraining the entire decoding layer may introduce too many parameters (head_num * last_hidden_size * vocab_size). MEDUSA -1: MEDUSA is directly fine-tuned on top of a frozen backbone LLM , enabling lossless inference acceleration .", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/FasterDecoding/Medusa", "content": "Medusa is a simple framework that democratizes the acceleration techniques for LLM generation with multiple decoding heads . Medusa -1 on Vicuna-7b. We aim to tackle the three pain points of popular acceleration techniques like speculative decoding : •Requirement of a good draft model. •System complexity. •Inefficiency when using sampling-based generation. See full list on github.com See full list on github.com medusa /model/medusa_model.py is the key file for Medusa . It contains the MedusaModel class, which is a wrapper of the original model and the new heads . This class also has an implementation of a streaming generation method. If you want to dive into the details of Medusa , this is the place to start. We also provide some illustrative notebooks in notebooks/ to help you understand the codebase. See full list on github.com We are super excited to see that Medusa has been adopted by many open-source projects. Here is an (incomplete) list: •TensorRT- LLM •TGI •RTP- LLM See full list on github.com We welcome community contributions to Medusa . If you have an idea for how to improve it, please open an issue to discuss it with us. When submitting a pull request, please ensure that your changes are well-tested. Please split each major change into a separate pull request. We also have a Roadmap summarizing our future plans for Medusa . Don't hesit... See full list on github.com This codebase is influenced by remarkable projects from the LLM community, including FastChat, TinyChat, vllm, axolotl. This project is supported by Together AI, MyShell AI, Chai AI. See full list on github.com Jul 1, 2025 · Today, I’m introducing a classic multi - head decoding architecture called Medusa . Medusa , inspired by the mythological Greek figure also known as the “snake-haired woman,” has each decoding head metaphorically representing a snake. The architecture mirrors this imagery with multiple decoding heads . Medusa seeks to improve model inference speed by adding additional decoding heads to existing LLMs . These additional heads enable multi-token prediction and can be trained with the backbone LLM frozen, or in conjunction with the original LLM. Read the paper: https://arxiv.org/abs/2401.10774. Specifically, we introduce MEDUSA , a method that enhances LLM inference by integrating additional decoding heads to concurrently predict multiple tokens. These heads are fine-tuned in a parameter-eficient manner and can be added to any existing model. Can Medusa augment LLM inference by adding additional decoding heads? In this paper, we present MEDUSA, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, MEDUSA constructs multiple candidate continuations and verifies them simultaneously in each decoding step. How does Medusa augment LLM inference? In this paper, we present Medusa, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, Medusa constructs multiple candidate continuations and verifies them simultaneously in each decoding step. Does Medusa speed up speculative decoding? The new results show a 2.2-3.6x speedup over the original model on a range of LLMs. Medusa is a simple framework that democratizes the acceleration techniques for LLM generation with multiple decoding heads . Medusa -1 on Vicuna-7b. We aim to tackle the three pain points of popular acceleration techniques like speculative decoding : How does Medusa reduce the number of decoding steps required? By leveraging parallel processing , Medusa substantially reduces the number of decoding steps required. We present two levels of fine-tuning procedures for Medusa to meet the needs of different use cases: Medusa-1: Medusa is directly fine-tuned on top of a frozen backbone LLM, enabling lossless inference acceleration. What's the difference between Medusa 2 & LLM? MEDUSA -2: MEDUSA is fine-tuned together with the backbone LLM , enabling better prediction accuracy of MEDUSA heads and higher speedup but needing a special training recipe that preserves the model's capabilities. What are the components of Medusa? The two most crucial components in Medusa are the multiple decoding heads and the tree-based attention mechanism . The decoding head structure is shown below: In practice, retraining the entire decoding layer may introduce too many parameters (head_num * last_hidden_size * vocab_size). MEDUSA -1: MEDUSA is directly fine-tuned on top of a frozen backbone LLM , enabling lossless inference acceleration ."}
+{"idx": 3, "title": "[Paper Reading] Medusa: Simple LLM Inference Acceleration ...", "date": "", "ddg_snippet": "Jul 1, 2025 · Today, I’m introducing a classic multi - head decoding architecture called Medusa . Medusa , inspired by the mythological Greek figure also known as the “snake-haired woman,” has each decoding head metaphorically representing a snake. The architecture mirrors this imagery with multiple decoding heads .", "subpage_snippet": "", "source": "clay-atlas.com", "link": "https://clay-atlas.com/us/blog/2025/07/01/en-medusa-simple-llm-inference-acceleration-framework-with-multiple-decoding-heads/", "content": "Jul 1, 2025 · Today, I’m introducing a classic multi - head decoding architecture called Medusa . Medusa , inspired by the mythological Greek figure also known as the “snake-haired woman,” has each decoding head metaphorically representing a snake. The architecture mirrors this imagery with multiple decoding heads ."}
+{"idx": 4, "title": "Medusa: Multiple Decoding Heads for Faster LLM Inference", "date": "", "ddg_snippet": "Medusa seeks to improve model inference speed by adding additional decoding heads to existing LLMs . These additional heads enable multi-token prediction and can be trained with the backbone LLM frozen, or in conjunction with the original LLM. Read the paper: https://arxiv.org/abs/2401.10774.", "subpage_snippet": "", "source": "mlscrapbook.substack.com", "link": "https://mlscrapbook.substack.com/p/medusa-multiple-decoding-heads-for", "content": "Medusa seeks to improve model inference speed by adding additional decoding heads to existing LLMs . These additional heads enable multi-token prediction and can be trained with the backbone LLM frozen, or in conjunction with the original LLM. Read the paper: https://arxiv.org/abs/2401.10774."}
+{"idx": 5, "title": "Medusa: Simple LLM Inference Acceleration Framework with ...", "date": "", "ddg_snippet": "Specifically, we introduce MEDUSA , a method that enhances LLM inference by integrating additional decoding heads to concurrently predict multiple tokens. These heads are fine-tuned in a parameter-eficient manner and can be added to any existing model.", "subpage_snippet": "", "source": "raw.githubusercontent.com", "link": "https://raw.githubusercontent.com/mlresearch/v235/main/assets/cai24b/cai24b.pdf", "content": "Specifically, we introduce MEDUSA , a method that enhances LLM inference by integrating additional decoding heads to concurrently predict multiple tokens. These heads are fine-tuned in a parameter-eficient manner and can be added to any existing model."}
+{"idx": 6, "title": "Abstract MEDUSA: Simple LLM Inference Acceleration Framework ...", "date": "", "ddg_snippet": "MEDUSA -1: MEDUSA is directly fine-tuned on top of a frozen backbone LLM , enabling lossless inference acceleration .", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2401.10774v1", "content": "MEDUSA -1: MEDUSA is directly fine-tuned on top of a frozen backbone LLM , enabling lossless inference acceleration ."}
+{"idx": 7, "title": "Medusa: Simple LLM Inference Acceleration Framework ...", "date": "", "ddg_snippet": "by T Cai · 2024 · Cited by 348 — Abstract. Large Language Models (LLMs) employ auto- regressive decoding that requires sequential com- putation, with each step reliant on the previous.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2401.10774", "content": "by T Cai · 2024 · Cited by 348 — Abstract. Large Language Models (LLMs) employ auto- regressive decoding that requires sequential com- putation, with each step reliant on the previous."}
+{"idx": 8, "title": "MEDUSA: Simple LLM Inference Acceleration Framework ...", "date": "", "ddg_snippet": "by T Cai · 2024 · Cited by 348 — In this paper, we present MEDUSA , an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in ...", "subpage_snippet": "", "source": "experts.illinois.edu", "link": "https://experts.illinois.edu/en/publications/medusa-simple-llm-inference-acceleration-framework-with-multiple-", "content": "by T Cai · 2024 · Cited by 348 — In this paper, we present MEDUSA , an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in ..."}
+{"idx": 9, "title": "Exploring Medusa and Multi-Token Prediction", "date": "", "ddg_snippet": "This blog post will go into detail on the “ MEDUSA : Simple LLM Inference Acceleration Framework with Multiple Decoding Heads ” paper.", "subpage_snippet": "", "source": "medium.com", "link": "https://medium.com/data-science/exploring-medusa-and-multi-token-prediction-de7f8312e4a7", "content": "This blog post will go into detail on the “ MEDUSA : Simple LLM Inference Acceleration Framework with Multiple Decoding Heads ” paper."}
diff --git a/data/sampled_jsons/MultiPDENet_Physics_Block_PDE-embedded_learning_Runge-Kutta_integrator_methodology_year_2024.jsonl b/data/sampled_jsons/MultiPDENet_Physics_Block_PDE-embedded_learning_Runge-Kutta_integrator_methodology_year_2024.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..93c62363130d1a5b2b130eed788266c9e2ea694b
--- /dev/null
+++ b/data/sampled_jsons/MultiPDENet_Physics_Block_PDE-embedded_learning_Runge-Kutta_integrator_methodology_year_2024.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "[2501.15987] MultiPDENet: PDE-embedded Learning with Multi ... PDE-constrained Learning with Multi-time-stepping for ... [PDF] MultiPDENet: PDE-embedded Learning with Multi-time ... GitHub - ZichaoLong/PDE-Net: PDE-Net: Learning PDEs from Data MultiPDENet: PDE-embedded Learning with Multi-time-stepping ... High-Order Implicit-Explicit Multi-Block Time-stepping Method ... Modeling the dynamics of PDE systems with physics-constrained ...", "date": "", "ddg_snippet": "Jan 27, 2025 · A Physics Block with a 4th-order Runge-Kutta integrator at the fine time scale is established that embeds the structure of PDEs to guide the prediction. Sep 27, 2024 · A physics block with a 4th-order Runge-Kutta integrator at the fine time scale is established that embeds the structure of PDEs to guide the prediction. Jan 27, 2025 · A PDE-embedded network with multiscale time stepping ( MultiPDENet ), which fuses the scheme of numerical methods and machine learning , for accelerated simulation of flows and achieves the state-of-the-art performance compared with other neural baseline models, also with clear speedup compared to classical numerical methods . Solving partial differential equations (PDEs) by numerical methods meet ... PDE -Net: Learning PDEs from Data. Contribute to ZichaoLong/ PDE -Net development by creating an account on GitHub. Technical Explanation The network architecture uses a multi-time-stepping approach with three key components: a physics -informed neural network, a correction module, and a multi-step integration scheme. The system employs residual learning to capture fine-scale flow features while using larger time steps. The system of ODEs are integrated in time using a high-order Runge-Kutta method (see section C). The way in which the boundary conditions are imposed play a vital role when solving PDEs and will be discussed in section B. Feb 15, 2020 · Finally, we compare the prediction computational cost of this surrogate model with both FEM and finite difference method (FDM) using fourth-order Runge-Kutta time integration in Table 2.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2501.15987", "content": "Jan 27, 2025 · A Physics Block with a 4th-order Runge-Kutta integrator at the fine time scale is established that embeds the structure of PDEs to guide the prediction. Sep 27, 2024 · A physics block with a 4th-order Runge-Kutta integrator at the fine time scale is established that embeds the structure of PDEs to guide the prediction. Jan 27, 2025 · A PDE-embedded network with multiscale time stepping ( MultiPDENet ), which fuses the scheme of numerical methods and machine learning , for accelerated simulation of flows and achieves the state-of-the-art performance compared with other neural baseline models, also with clear speedup compared to classical numerical methods . Solving partial differential equations (PDEs) by numerical methods meet ... PDE -Net: Learning PDEs from Data. Contribute to ZichaoLong/ PDE -Net development by creating an account on GitHub. Technical Explanation The network architecture uses a multi-time-stepping approach with three key components: a physics -informed neural network, a correction module, and a multi-step integration scheme. The system employs residual learning to capture fine-scale flow features while using larger time steps. The system of ODEs are integrated in time using a high-order Runge-Kutta method (see section C). The way in which the boundary conditions are imposed play a vital role when solving PDEs and will be discussed in section B. Feb 15, 2020 · Finally, we compare the prediction computational cost of this surrogate model with both FEM and finite difference method (FDM) using fourth-order Runge-Kutta time integration in Table 2."}
+{"idx": 1, "title": "PDE-constrained Learning with Multi-time-stepping for ...", "date": "", "ddg_snippet": "Sep 27, 2024 · A physics block with a 4th-order Runge-Kutta integrator at the fine time scale is established that embeds the structure of PDEs to guide the prediction.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=stcN89QGfL", "content": "Sep 27, 2024 · A physics block with a 4th-order Runge-Kutta integrator at the fine time scale is established that embeds the structure of PDEs to guide the prediction."}
+{"idx": 2, "title": "[PDF] MultiPDENet: PDE-embedded Learning with Multi-time ...", "date": "", "ddg_snippet": "Jan 27, 2025 · A PDE-embedded network with multiscale time stepping ( MultiPDENet ), which fuses the scheme of numerical methods and machine learning , for accelerated simulation of flows and achieves the state-of-the-art performance compared with other neural baseline models, also with clear speedup compared to classical numerical methods . Solving partial differential equations (PDEs) by numerical methods meet ...", "subpage_snippet": "", "source": "www.semanticscholar.org", "link": "https://www.semanticscholar.org/paper/MultiPDENet:-PDE-embedded-Learning-with-for-Flow-Wang-Mi/6aee4adf8e7489f251995859a5f0432a2c60bb82", "content": "Jan 27, 2025 · A PDE-embedded network with multiscale time stepping ( MultiPDENet ), which fuses the scheme of numerical methods and machine learning , for accelerated simulation of flows and achieves the state-of-the-art performance compared with other neural baseline models, also with clear speedup compared to classical numerical methods . Solving partial differential equations (PDEs) by numerical methods meet ..."}
+{"idx": 3, "title": "GitHub - ZichaoLong/PDE-Net: PDE-Net: Learning PDEs from Data", "date": "", "ddg_snippet": "PDE -Net: Learning PDEs from Data. Contribute to ZichaoLong/ PDE -Net development by creating an account on GitHub.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/ZichaoLong/PDE-Net", "content": "PDE -Net: Learning PDEs from Data. Contribute to ZichaoLong/ PDE -Net development by creating an account on GitHub."}
+{"idx": 4, "title": "MultiPDENet: PDE-embedded Learning with Multi-time-stepping ...", "date": "", "ddg_snippet": "Technical Explanation The network architecture uses a multi-time-stepping approach with three key components: a physics -informed neural network, a correction module, and a multi-step integration scheme. The system employs residual learning to capture fine-scale flow features while using larger time steps.", "subpage_snippet": "", "source": "www.aimodels.fyi", "link": "https://www.aimodels.fyi/papers/arxiv/multipdenet-pde-embedded-learning-multi-time-stepping", "content": "Technical Explanation The network architecture uses a multi-time-stepping approach with three key components: a physics -informed neural network, a correction module, and a multi-step integration scheme. The system employs residual learning to capture fine-scale flow features while using larger time steps."}
+{"idx": 5, "title": "High-Order Implicit-Explicit Multi-Block Time-stepping Method ...", "date": "", "ddg_snippet": "The system of ODEs are integrated in time using a high-order Runge-Kutta method (see section C). The way in which the boundary conditions are imposed play a vital role when solving PDEs and will be discussed in section B.", "subpage_snippet": "", "source": "ntrs.nasa.gov", "link": "https://ntrs.nasa.gov/api/citations/20140004064/downloads/20140004064.pdf", "content": "The system of ODEs are integrated in time using a high-order Runge-Kutta method (see section C). The way in which the boundary conditions are imposed play a vital role when solving PDEs and will be discussed in section B."}
+{"idx": 6, "title": "Modeling the dynamics of PDE systems with physics-constrained ...", "date": "", "ddg_snippet": "Feb 15, 2020 · Finally, we compare the prediction computational cost of this surrogate model with both FEM and finite difference method (FDM) using fourth-order Runge-Kutta time integration in Table 2.", "subpage_snippet": "", "source": "www.sciencedirect.com", "link": "https://www.sciencedirect.com/science/article/pii/S0021999119307612", "content": "Feb 15, 2020 · Finally, we compare the prediction computational cost of this surrogate model with both FEM and finite difference method (FDM) using fourth-order Runge-Kutta time integration in Table 2."}
+{"idx": 7, "title": "MultiPDENet: PDE-embedded Learning with Multi-time- ...", "date": "", "ddg_snippet": "by Q Wang — The paper introduces MultiPDENet , a PDE-embedded neural network with multiscale time stepping to accelerate flow simulations by integrating numerical methods ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=D1gs8QT74m", "content": "by Q Wang — The paper introduces MultiPDENet , a PDE-embedded neural network with multiscale time stepping to accelerate flow simulations by integrating numerical methods ..."}
+{"idx": 8, "title": "MultiPDENet: PDE-embedded Learning with Multi-time ...", "date": "", "ddg_snippet": "15 Jul 2025 — A Physics Block with a 4th-order Runge - Kutta integrator at the fine time scale is established that embeds the structure of PDEs to guide the ...", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/poster/46029", "content": "15 Jul 2025 — A Physics Block with a 4th-order Runge - Kutta integrator at the fine time scale is established that embeds the structure of PDEs to guide the ..."}
+{"idx": 9, "title": "MultiPDENet: PDE-embedded Learning with Multi-time ...", "date": "", "ddg_snippet": "27 Jan 2025 — A Physics Block with a 4th-order Runge - Kutta integrator at the fine time scale is established that embeds the structure of PDEs to guide the ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2501.15987v1", "content": "27 Jan 2025 — A Physics Block with a 4th-order Runge - Kutta integrator at the fine time scale is established that embeds the structure of PDEs to guide the ..."}
diff --git a/data/sampled_jsons/OmniBench_OpenReview_PDF_Section_5.1_NVIDIA_A100.jsonl b/data/sampled_jsons/OmniBench_OpenReview_PDF_Section_5.1_NVIDIA_A100.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..cfd3884024d4cc2871c41363db28e7ea022f69ee
--- /dev/null
+++ b/data/sampled_jsons/OmniBench_OpenReview_PDF_Section_5.1_NVIDIA_A100.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "OMNIBENCH: TOWARDS THE FUTURE OF UNIVERSAL OMNI-LANGUAGE ... - OpenReview", "date": "", "ddg_snippet": "Recent advancements in multimodal large language models (MLLMs) have aimed to integrate and interpret data across diverse modalities. However, the capac-ity of these models to concurrently process and reason about multiple modalities remains underexplored, partly due to the lack of comprehensive modality-wise benchmarks. We introduce OmniBench , a novel benchmark designed to rigor-ously ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=Rc8z5wLzBF", "content": "Recent advancements in multimodal large language models (MLLMs) have aimed to integrate and interpret data across diverse modalities. However, the capac-ity of these models to concurrently process and reason about multiple modalities remains underexplored, partly due to the lack of comprehensive modality-wise benchmarks. We introduce OmniBench , a novel benchmark designed to rigor-ously ..."}
+{"idx": 1, "title": "OmniBench: Towards The Future of Universal Omni-Language Models", "date": "", "ddg_snippet": "Impact Statement This paper introduces OmniBench , a crucial step towards developing truly multimodal AI. By rigorously evaluating models on their ability to integrate visual, acoustic, and textual information, OmniBench exposes critical limitations in current approaches and highlights the need for dedicated research in tri-modal reasoning.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2409.15272v4", "content": "Impact Statement This paper introduces OmniBench , a crucial step towards developing truly multimodal AI. By rigorously evaluating models on their ability to integrate visual, acoustic, and textual information, OmniBench exposes critical limitations in current approaches and highlights the need for dedicated research in tri-modal reasoning."}
+{"idx": 2, "title": "PDF NVIDIA A100 Tensor Core GPU Architecture", "date": "", "ddg_snippet": "The NVIDIA A100 GPU based on NVIDIA Ampere architecture is engineered to provide as much AI and HPC computing power as possible from its many new architectural features and optimizations.", "subpage_snippet": "", "source": "images.nvidia.com", "link": "https://images.nvidia.com/aem-dam/en-zz/Solutions/data-center/nvidia-ampere-architecture-whitepaper.pdf", "content": "The NVIDIA A100 GPU based on NVIDIA Ampere architecture is engineered to provide as much AI and HPC computing power as possible from its many new architectural features and optimizations."}
+{"idx": 3, "title": "NVIDIA A100 GPU Benchmarks for Deep Learning - Lambda Labs", "date": "", "ddg_snippet": "Benchmarks for ResNet-152, Inception v3, Inception v4, VGG-16, AlexNet, SSD300, and ResNet-50 using the NVIDIA A100 GPU and DGX A100 server.", "subpage_snippet": "", "source": "lambda.ai", "link": "https://lambda.ai/blog/nvidia-a100-gpu-deep-learning-benchmarks-and-architectural-overview", "content": "Benchmarks for ResNet-152, Inception v3, Inception v4, VGG-16, AlexNet, SSD300, and ResNet-50 using the NVIDIA A100 GPU and DGX A100 server."}
+{"idx": 4, "title": "OmniBench: Towards The Future of Universal Omni-Language Models", "date": "", "ddg_snippet": "This paper presents OmniBench , a multi-modal benchmark developed to evaluate the capacity of large multimodal language models (MLLMs) to process and reason across visual, auditory, and textual modalities. In this framework, the authors classify these systems as omni -language models (OLMs) and set a unique requirement for accurate responses that reflect an integrated understanding across all ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=Rc8z5wLzBF", "content": "This paper presents OmniBench , a multi-modal benchmark developed to evaluate the capacity of large multimodal language models (MLLMs) to process and reason across visual, auditory, and textual modalities. In this framework, the authors classify these systems as omni -language models (OLMs) and set a unique requirement for accurate responses that reflect an integrated understanding across all ..."}
+{"idx": 5, "title": "PDF Benchmarking the NVIDIA A100 Graphics Processing Unit for High ...", "date": "", "ddg_snippet": "However, these ML workloads also benefit from increased tensor core capabilities in the V100 and A100 GPUs, yielding a 3.5x speedup using a mixed (single + half) precision strategy for floating point operations. While the performance gap between GPUs and CPUs remains moderate (3x to 8x) for high-performance computing applications, these new hardware features of recent GPU generations give 50x ...", "subpage_snippet": "", "source": "craigulmer.com", "link": "https://craigulmer.com/data/2021/SAND2021-1220_uur.pdf", "content": "However, these ML workloads also benefit from increased tensor core capabilities in the V100 and A100 GPUs, yielding a 3.5x speedup using a mixed (single + half) precision strategy for floating point operations. While the performance gap between GPUs and CPUs remains moderate (3x to 8x) for high-performance computing applications, these new hardware features of recent GPU generations give 50x ..."}
+{"idx": 6, "title": "OmniBench: Towards The Future of Universal Omni-Language Models", "date": "", "ddg_snippet": "These results underscore the importance of OmniBench as a tool for identifying areas of improvement and guiding research in multimodal systems. In the following sections , we detail the data collection protocol of OmniBench , present our evaluation results on current state-of-the-art MLLMs, and discuss the implications of our findings for the future of research and development. Through OmniBench ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2409.15272v1", "content": "These results underscore the importance of OmniBench as a tool for identifying areas of improvement and guiding research in multimodal systems. In the following sections , we detail the data collection protocol of OmniBench , present our evaluation results on current state-of-the-art MLLMs, and discuss the implications of our findings for the future of research and development. Through OmniBench ..."}
+{"idx": 7, "title": "PDF Evaluating the Performance of NVIDIA's A100 Ampere GPU for Sparse and ...", "date": "", "ddg_snippet": "For the performance assessment, in Section II we first benchmark the bandwidth of the A100 GPU for memory-bound vector operations and compare against NVIDIA's A100 predecessor, the V100 GPU.", "subpage_snippet": "", "source": "icl.utk.edu", "link": "https://icl.utk.edu/files/publications/2020/icl-utk-1441-2020.pdf", "content": "For the performance assessment, in Section II we first benchmark the bandwidth of the A100 GPU for memory-bound vector operations and compare against NVIDIA's A100 predecessor, the V100 GPU."}
+{"idx": 8, "title": "PDF Nvidia Dgx A100", "date": "", "ddg_snippet": "The NVIDIA DGX OS software supports the ability to manage self-encrypting drives (SEDs), including setting an Authentication Key for locking and unlocking the drives on NVIDIA DGXTM A100 systems.", "subpage_snippet": "", "source": "docs.nvidia.com", "link": "https://docs.nvidia.com/dgx/pdf/dgxa100-user-guide.pdf", "content": "The NVIDIA DGX OS software supports the ability to manage self-encrypting drives (SEDs), including setting an Authentication Key for locking and unlocking the drives on NVIDIA DGXTM A100 systems."}
+{"idx": 9, "title": "Nvidia A100 | Nvidia", "date": "", "ddg_snippet": "NVIDIA A100 Tensor Core GPU delivers unprecedented acceleration at every scale to power the world's highest-performing elastic data centers for AI, data analytics, and HPC. Powered by the NVIDIA Ampere Architecture, A100 is the engine of the NVIDIA data center platform. A100 provides up to 20X higher performance over the prior generation and can be partitioned into seven GPU instances to ...", "subpage_snippet": "", "source": "www.nvidia.com", "link": "https://www.nvidia.com/en-us/data-center/a100/", "content": "NVIDIA A100 Tensor Core GPU delivers unprecedented acceleration at every scale to power the world's highest-performing elastic data centers for AI, data analytics, and HPC. Powered by the NVIDIA Ampere Architecture, A100 is the engine of the NVIDIA data center platform. A100 provides up to 20X higher performance over the prior generation and can be partitioned into seven GPU instances to ..."}
diff --git a/data/sampled_jsons/On_the_Role_of_Attention_Heads_in_Large_Language_Model_Safety_experimental_results_ASR_Llama-2-7b-ch.jsonl b/data/sampled_jsons/On_the_Role_of_Attention_Heads_in_Large_Language_Model_Safety_experimental_results_ASR_Llama-2-7b-ch.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..bc89e8b4a1ef67db2a320085718edf3006bf06c0
--- /dev/null
+++ b/data/sampled_jsons/On_the_Role_of_Attention_Heads_in_Large_Language_Model_Safety_experimental_results_ASR_Llama-2-7b-ch.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Safeguarding Large Reasoning Models with Inference-time ...", "date": "", "ddg_snippet": "by Y Wang · 2025 — By monitoring models ' internal attention behaviors, ReasoningGuard locates the turning point of reasoning behaviors, and timely injects safety ...", "subpage_snippet": "", "source": "www.arxiv.org", "link": "https://www.arxiv.org/pdf/2508.04204", "content": "by Y Wang · 2025 — By monitoring models ' internal attention behaviors, ReasoningGuard locates the turning point of reasoning behaviors, and timely injects safety ..."}
+{"idx": 1, "title": "Revealing the Hidden Weakness in Aligned LLMs' Refusal ...", "date": "", "ddg_snippet": "by J Yu — We compare the atten- tion for eos tokens (as used in BOOST) versus GCG-generated tokens within the Llama - 2 - 7b - chat model (-10th layer, 0-th head ). Following ...", "subpage_snippet": "", "source": "www.usenix.org", "link": "https://www.usenix.org/system/files/usenixsecurity25-yu-jiahao.pdf", "content": "by J Yu — We compare the atten- tion for eos tokens (as used in BOOST) versus GCG-generated tokens within the Llama - 2 - 7b - chat model (-10th layer, 0-th head ). Following ..."}
+{"idx": 2, "title": "LLMs know their vulnerabilities: Uncover Safety Gaps ...", "date": "", "ddg_snippet": "by Q Ren · 2025 · Cited by 2 — Safety concerns in large language models . (LLMs) have gained significant attention due to their exposure to potentially harmful data . 23 pages", "subpage_snippet": "", "source": "aclanthology.org", "link": "https://aclanthology.org/2025.acl-long.1207.pdf", "content": "by Q Ren · 2025 · Cited by 2 — Safety concerns in large language models . (LLMs) have gained significant attention due to their exposure to potentially harmful data . 23 pages"}
+{"idx": 3, "title": "Daily Papers", "date": "", "ddg_snippet": "Ablating a single safety head allows aligned model (e.g., Llama - 2 - 7b - chat ) to respond to 16 times more harmful queries, while only modifying 0.006% of the ...", "subpage_snippet": "", "source": "huggingface.co", "link": "https://huggingface.co/papers?q=safety+guard", "content": "Ablating a single safety head allows aligned model (e.g., Llama - 2 - 7b - chat ) to respond to 16 times more harmful queries, while only modifying 0.006% of the ..."}
+{"idx": 4, "title": "Daily Papers", "date": "", "ddg_snippet": "On the Role of Attention Heads in Large Language Model Safety · Large language models (LLMs) achieve state-of-the-art performance on multiple language tasks ...", "subpage_snippet": "", "source": "huggingface.co", "link": "https://huggingface.co/papers?q=multi-turn+safety+training+dataset", "content": "On the Role of Attention Heads in Large Language Model Safety · Large language models (LLMs) achieve state-of-the-art performance on multiple language tasks ..."}
+{"idx": 5, "title": "TwinBreak: Jailbreaking LLM Security Alignments based ...", "date": "", "ddg_snippet": "9 Jun 2025 — TwinBreak is the first method to analyze intermediate outputs from prompts with high structural and content similarity to isolate safety parameters.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2506.07596v1", "content": "9 Jun 2025 — TwinBreak is the first method to analyze intermediate outputs from prompts with high structural and content similarity to isolate safety parameters."}
+{"idx": 6, "title": "Don't Say No: Jailbreaking LLM by Suppressing Refusal", "date": "", "ddg_snippet": "by Y Zhou · 2025 · Cited by 45 — These results suggest that learning-based methods effectively exploit alignment vulnerabil- ities in LLMs, making jailbreak attacks context-.", "subpage_snippet": "", "source": "aclanthology.org", "link": "https://aclanthology.org/2025.findings-acl.1294.pdf", "content": "by Y Zhou · 2025 · Cited by 45 — These results suggest that learning-based methods effectively exploit alignment vulnerabil- ities in LLMs, making jailbreak attacks context-."}
+{"idx": 7, "title": "Blackbox LLM Vulnerabilities", "date": "", "ddg_snippet": "Large Language Models (LLMs) are vulnerable to activation steering attacks that bypass safety and privacy mechanisms. By manipulating internal attention head ...", "subpage_snippet": "", "source": "www.promptfoo.dev", "link": "https://www.promptfoo.dev/lm-security-db/tag/blackbox", "content": "Large Language Models (LLMs) are vulnerable to activation steering attacks that bypass safety and privacy mechanisms. By manipulating internal attention head ..."}
+{"idx": 8, "title": "© 2024 Mantas Mazeika - IDEALS", "date": "", "ddg_snippet": "by M Mazeika · 2024 — In Figure 2.5, we show that Zephyr 7B + R2D2 has the third lowest average ASR of all models , behind only Llama 2 7B Chat and Llama 2 13B. Chat. Compared to ...", "subpage_snippet": "", "source": "www.ideals.illinois.edu", "link": "https://www.ideals.illinois.edu/items/131475/bitstreams/436956/data.pdf", "content": "by M Mazeika · 2024 — In Figure 2.5, we show that Zephyr 7B + R2D2 has the third lowest average ASR of all models , behind only Llama 2 7B Chat and Llama 2 13B. Chat. Compared to ..."}
+{"idx": 9, "title": "Jailbreak LLM Vulnerabilities", "date": "", "ddg_snippet": "Large Language Models (LLMs) are vulnerable to activation steering attacks that bypass safety and privacy mechanisms. By manipulating internal attention head ...", "subpage_snippet": "", "source": "www.promptfoo.dev", "link": "https://www.promptfoo.dev/lm-security-db/tag/jailbreak", "content": "Large Language Models (LLMs) are vulnerable to activation steering attacks that bypass safety and privacy mechanisms. By manipulating internal attention head ..."}
diff --git a/data/sampled_jsons/OpenCLIP_framework_sigmoid_contrastive_learning_models_examples.jsonl b/data/sampled_jsons/OpenCLIP_framework_sigmoid_contrastive_learning_models_examples.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..c2b29a2d4698e5d58cc63e924e7c9849cf9efb85
--- /dev/null
+++ b/data/sampled_jsons/OpenCLIP_framework_sigmoid_contrastive_learning_models_examples.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "SigLIP Models | mlfoundations/open_clip | DeepWiki", "date": "", "ddg_snippet": "Apr 18, 2025 · SigLIP models represent an advanced variant of CLIP models in the OpenCLIP framework , featuring sigmoid -based contrastive learning and several architectural modifications.", "subpage_snippet": "", "source": "deepwiki.com", "link": "https://deepwiki.com/mlfoundations/open_clip/5.2-siglip-models", "content": "Apr 18, 2025 · SigLIP models represent an advanced variant of CLIP models in the OpenCLIP framework , featuring sigmoid -based contrastive learning and several architectural modifications."}
+{"idx": 1, "title": "GitHub - mlfoundations/open_clip: An open source ... SigLIP vs. CLIP: The Sigmoid Advantage | by heping_LU | Medium CLIP to SigLIP: Vision-Language Models with Contrastive Learning Inference_with_ (multilingual)_SigLIP,_a_better_CLIP_model ... GitHub - ramanakshay/clip: CLIP & SigLIP model training from ... SigLIP Models | mlfoundations/open_clip | DeepWiki SigLIP Models | mlfoundations/open_clip | DeepWiki SigLIP Models | mlfoundations/open_clip | DeepWiki SigLIP Models | mlfoundations/open_clip | DeepWiki GitHub - mlfoundations/open_ clip : An open source implementation of CLIP SigLIP vs. CLIP: The Sigmoid Advantage | by heping_LU open-clip-torch · PyPI", "date": "", "ddg_snippet": "[Paper] [Citations] [Clip Colab] [Coca Colab] Welcome to an open source implementation of OpenAI's CLIP ( Contrastive Language-Image Pre-training). Using this codebase, we have trained several models on a variety of data sources and compute budgets, ranging from small-scale experiments to larger runs including models trained on datasets such as LAION-400M, LAION-2B and DataComp-1B. Many of our models and their scaling properties are studied in detail in the paper reproducible scaling laws for contrastive language-image learning . Some of our best models and their zero-shot ImageNet-1k accuracy are shown below, along with the ViT-L model trained by OpenAI. We provide more details about our full collection of pretrained models here, and zero-shot results for 38 datasets here. Model cards with additional model specific details can be found on the Hugging Face Hub under the OpenCLIP library tag: https://huggingface.co/ models ?library=open_clip. If you found this repository useful, please consider citing. We welcome anyone to submit an issue or send an email if you have any other requests or suggestions. Note that portions of src/open_clip/ modelling and tokenizer code are adaptations of OpenAI's official repository. See full list on github.com Pretrained models We offer a simple model interface to instantiate both pre-trained and untrained models . To see which pretrained models are available, use the following code snippet. More details about our pretrained models are available here.You can find more about the models we support (e.g. number of parameters, FLOPs) in this table.NOTE: Many existing checkpoints use the QuickGELU activation from the original OpenAI models . This activation is actually less efficient than native torch.nn.GELU in recent versions of PyTorch. The model defaults are now nn.GELU, so one should use model definitions with -quickgelu postfix for the OpenCLIP pretrained weights. All OpenAI pretrained weights will always default to QuickGELU. One can also use the non -quickgelu model definitions with pretrained weights using QuickGELU but there will be an accuracy drop, for fine-tune that will likely vanish for longer runs. Future trained models will use nn.GELU. Loading models Models can be loaded with open_clip.create_model_and_transforms, as shown in the example below. The model name and corresponding pretrained keys are compatible with the outputs of open_clip.list_pretrained().The pretrained argument also accepts local paths, for example /path/to/my/b32.pt. You can also load checkpoints from huggingface this way. To do so, download the open_clip_pytorch_model.bin file (for example, https://huggingface.co/laion/CLIP-ViT-L-14-DataComp.XL-s13B-b90K/tree/main), and use pretrained=/path/to/open_clip_pytorch_model.bin. See full list on github.com This repository is focused on training CLIP models . To fine-tune a trained zero-shot model on a downstream classification task such as ImageNet, please see our other repository: WiSE-FT. The WiSE-FT repository contains code for our paper on Robust Fine-tuning of Zero-shot Models , in which we introduce a technique for fine-tuning zero-shot models wh... See full list on github.com Conceptual Captions See cc3m img2dataset example. YFCC and other datasets In addition to specifying the training data via CSV files as mentioned above, our codebase also supports webdataset, which is recommended for larger scale datasets. The expected format is a series of .tar files. Each of these .tar files should contain two files for each training example, one for the image and one for the corresponding text. Both files should have the same name but different extensions. For instance, shard_001.tar could contain files such as abc.jpg and abc.txt. You can learn more about webdataset at https://github.com/webdataset/webdataset. We use .tar files with 1,000 data points each, which we create using tarp.You can download the YFCC dataset from Multimedia Commons. Similar to OpenAI, we used a subset of YFCC to reach the aforementioned accuracy numbers. The indices of images in this subset are in OpenAI's CLIP repository. See full list on github.com Install We advise you first create a virtual environment with:You can then install openclip for training with pip install 'open_clip_torch[training]'. Sample single-process running code: Note: imagenet-val is the path to the validation set of ImageNet for zero-shot evaluation, not the training set! You can remove this argument if you do not want to perform zero-shot evaluation on ImageNet throughout training. Note that the val folder should contain subfolders. If it does not, please use this script. Multi-GPU and Beyond This code has been battle tested up to 1024 A100s and offers a variety of solutions for distributed training. We include native support for SLURM clusters.As the number of devices used to train increases, so does the space complexity of the the logit matrix. Using a naïve all-gather scheme, space complexity will be O(n^2). Instead, complexity may become effectively linear if the flags --gather-with-grad and --local-loss are used. This alteration results in one-to-one numerical results as the naïve method. See full list on github.com See full list on github.com We gratefully acknowledge the Gauss Centre for Supercomputing e.V. (www.gauss-centre.eu) for funding this part of work by providing computing time through the John von Neumann Institute for Computing (NIC) on the GCS Supercomputer JUWELS Booster at Jülich Supercomputing Centre (JSC). See full list on github.com Current development of this repository is led by Ross Wightman, Romain Beaumont, Cade Gordon, and Vaishaal Shankar. The original version of this repository is from a group of researchers at UW, Google, Stanford, Amazon, Columbia, and Berkeley. Gabriel Ilharco*, Mitchell Wortsman*, Nicholas Carlini, Rohan Taori, Achal Dave, Vaishaal Shankar, John Miller, Hongseok Namkoong, Hannaneh Hajishirzi, Ali Farhadi, Ludwig Schmidt Special thanks to Jong Wook Kim and Alec Radford for help with reproducing CLIP! See full list on github.com If you found this repository useful, please consider citing: See full list on github.com Sep 25, 2024 · Contrastive pre-training, using weakly supervised image-text pairs, has become the leading method for developing general computer vision models . This involves learning aligned representations for ... Aug 2, 2024 · Understand the changes from CLIP to SigLIP in vision-language models , highlighting the advantages of contrastive learning and how they enhances efficiency The sigmoid loss simultaneously allows further scaling up the batch size during pre-training (like 32k image-text pairs), while also performing better at smaller batch sizes. The model outperforms CLIP on both zero-shot image classification and image-text retrieval as shown below. Contrastive Language-Image Pre-training (CLIP) is a technique for training a pair of neural network models , one for image understanding and one for text understanding, using a contrastive objective. This project contains code to train CLIP on the MS-COCO Captions dataset. It also includes an implementation of SigLIP, which uses a sigmoid loss as the training objective. How do siglip models differ from standard Clip models? SigLIP models differ from standard CLIP in several important ways: Additionally, SigLIP models use a different tokenization approach with a \"canonicalize\" cleaning method for text input . Sources: src/open_clip/model_configs/ViT-B-16-SigLIP2.json src/open_clip/model_configs/ViT-B-16-SigLIP2-256.json What is siglip vs clip? SigLIP is an advanced variant of CLIP that uses a sigmoid-based contrastive loss function instead of the traditional softmax-based approach. This modification improves training stability and model performance, particularly for large-scale training. Sources: src/open_clip/model_configs/ViT-B-16-SigLIP2.json What are siglip models? SigLIP models represent an advanced variant of CLIP models in the OpenCLIP framework , featuring sigmoid-based contrastive learning and several architectural modifications. These models are available in multiple resolution variants and can be used with the standard OpenCLIP interfaces for training and inference. Which siglip model is best for training? The training configuration will need to specify the sigmoid-based loss instead of the standard softmax-based CLIP loss. Higher-resolution SigLIP models (384×384, 512×512) generally provide better performance but require more computational resources for both training and inference. Where can I find an open source implementation of clip? GitHub - mlfoundations/open_clip: An open source implementation of CLIP. Move src/clip -> src/open_clip, create setup.py + MANIFEST.in and tes… [Paper] [Citations] [Clip Colab] [Coca Colab] Welcome to an open source implementation of OpenAI's CLIP (Contrastive Language-Image Pre-training). Does siglip outperform clip? A final cross-device sum brings everything together. SigLIP outperforms CLIP at small batch sizes (e.g., 4–8k), but both reach saturation at 32k batch size despite claims that larger batches improve performance. LiT: SigLIP results, trained for 9B seen examples. Aug 6, 2025 · OpenCLIP [Paper] [Citations] [Clip Colab] [Coca Colab] Welcome to an open source implementation of OpenAI's CLIP ( Contrastive Language-Image Pre-training). Using this codebase, we have trained several models on a variety of data sources and compute budgets, ranging from small-scale experiments to larger runs including models trained on datasets such as LAION-400M, LAION-2B and DataComp-1B ...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/mlfoundations/open_clip", "content": "[Paper] [Citations] [Clip Colab] [Coca Colab] Welcome to an open source implementation of OpenAI's CLIP ( Contrastive Language-Image Pre-training). Using this codebase, we have trained several models on a variety of data sources and compute budgets, ranging from small-scale experiments to larger runs including models trained on datasets such as LAION-400M, LAION-2B and DataComp-1B. Many of our models and their scaling properties are studied in detail in the paper reproducible scaling laws for contrastive language-image learning . Some of our best models and their zero-shot ImageNet-1k accuracy are shown below, along with the ViT-L model trained by OpenAI. We provide more details about our full collection of pretrained models here, and zero-shot results for 38 datasets here. Model cards with additional model specific details can be found on the Hugging Face Hub under the OpenCLIP library tag: https://huggingface.co/ models ?library=open_clip. If you found this repository useful, please consider citing. We welcome anyone to submit an issue or send an email if you have any other requests or suggestions. Note that portions of src/open_clip/ modelling and tokenizer code are adaptations of OpenAI's official repository. See full list on github.com Pretrained models We offer a simple model interface to instantiate both pre-trained and untrained models . To see which pretrained models are available, use the following code snippet. More details about our pretrained models are available here.You can find more about the models we support (e.g. number of parameters, FLOPs) in this table.NOTE: Many existing checkpoints use the QuickGELU activation from the original OpenAI models . This activation is actually less efficient than native torch.nn.GELU in recent versions of PyTorch. The model defaults are now nn.GELU, so one should use model definitions with -quickgelu postfix for the OpenCLIP pretrained weights. All OpenAI pretrained weights will always default to QuickGELU. One can also use the non -quickgelu model definitions with pretrained weights using QuickGELU but there will be an accuracy drop, for fine-tune that will likely vanish for longer runs. Future trained models will use nn.GELU. Loading models Models can be loaded with open_clip.create_model_and_transforms, as shown in the example below. The model name and corresponding pretrained keys are compatible with the outputs of open_clip.list_pretrained().The pretrained argument also accepts local paths, for example /path/to/my/b32.pt. You can also load checkpoints from huggingface this way. To do so, download the open_clip_pytorch_model.bin file (for example, https://huggingface.co/laion/CLIP-ViT-L-14-DataComp.XL-s13B-b90K/tree/main), and use pretrained=/path/to/open_clip_pytorch_model.bin. See full list on github.com This repository is focused on training CLIP models . To fine-tune a trained zero-shot model on a downstream classification task such as ImageNet, please see our other repository: WiSE-FT. The WiSE-FT repository contains code for our paper on Robust Fine-tuning of Zero-shot Models , in which we introduce a technique for fine-tuning zero-shot models wh... See full list on github.com Conceptual Captions See cc3m img2dataset example. YFCC and other datasets In addition to specifying the training data via CSV files as mentioned above, our codebase also supports webdataset, which is recommended for larger scale datasets. The expected format is a series of .tar files. Each of these .tar files should contain two files for each training example, one for the image and one for the corresponding text. Both files should have the same name but different extensions. For instance, shard_001.tar could contain files such as abc.jpg and abc.txt. You can learn more about webdataset at https://github.com/webdataset/webdataset. We use .tar files with 1,000 data points each, which we create using tarp.You can download the YFCC dataset from Multimedia Commons. Similar to OpenAI, we used a subset of YFCC to reach the aforementioned accuracy numbers. The indices of images in this subset are in OpenAI's CLIP repository. See full list on github.com Install We advise you first create a virtual environment with:You can then install openclip for training with pip install 'open_clip_torch[training]'. Sample single-process running code: Note: imagenet-val is the path to the validation set of ImageNet for zero-shot evaluation, not the training set! You can remove this argument if you do not want to perform zero-shot evaluation on ImageNet throughout training. Note that the val folder should contain subfolders. If it does not, please use this script. Multi-GPU and Beyond This code has been battle tested up to 1024 A100s and offers a variety of solutions for distributed training. We include native support for SLURM clusters.As the number of devices used to train increases, so does the space complexity of the the logit matrix. Using a naïve all-gather scheme, space complexity will be O(n^2). Instead, complexity may become effectively linear if the flags --gather-with-grad and --local-loss are used. This alteration results in one-to-one numerical results as the naïve method. See full list on github.com See full list on github.com We gratefully acknowledge the Gauss Centre for Supercomputing e.V. (www.gauss-centre.eu) for funding this part of work by providing computing time through the John von Neumann Institute for Computing (NIC) on the GCS Supercomputer JUWELS Booster at Jülich Supercomputing Centre (JSC). See full list on github.com Current development of this repository is led by Ross Wightman, Romain Beaumont, Cade Gordon, and Vaishaal Shankar. The original version of this repository is from a group of researchers at UW, Google, Stanford, Amazon, Columbia, and Berkeley. Gabriel Ilharco*, Mitchell Wortsman*, Nicholas Carlini, Rohan Taori, Achal Dave, Vaishaal Shankar, John Miller, Hongseok Namkoong, Hannaneh Hajishirzi, Ali Farhadi, Ludwig Schmidt Special thanks to Jong Wook Kim and Alec Radford for help with reproducing CLIP! See full list on github.com If you found this repository useful, please consider citing: See full list on github.com Sep 25, 2024 · Contrastive pre-training, using weakly supervised image-text pairs, has become the leading method for developing general computer vision models . This involves learning aligned representations for ... Aug 2, 2024 · Understand the changes from CLIP to SigLIP in vision-language models , highlighting the advantages of contrastive learning and how they enhances efficiency The sigmoid loss simultaneously allows further scaling up the batch size during pre-training (like 32k image-text pairs), while also performing better at smaller batch sizes. The model outperforms CLIP on both zero-shot image classification and image-text retrieval as shown below. Contrastive Language-Image Pre-training (CLIP) is a technique for training a pair of neural network models , one for image understanding and one for text understanding, using a contrastive objective. This project contains code to train CLIP on the MS-COCO Captions dataset. It also includes an implementation of SigLIP, which uses a sigmoid loss as the training objective. How do siglip models differ from standard Clip models? SigLIP models differ from standard CLIP in several important ways: Additionally, SigLIP models use a different tokenization approach with a \"canonicalize\" cleaning method for text input . Sources: src/open_clip/model_configs/ViT-B-16-SigLIP2.json src/open_clip/model_configs/ViT-B-16-SigLIP2-256.json What is siglip vs clip? SigLIP is an advanced variant of CLIP that uses a sigmoid-based contrastive loss function instead of the traditional softmax-based approach. This modification improves training stability and model performance, particularly for large-scale training. Sources: src/open_clip/model_configs/ViT-B-16-SigLIP2.json What are siglip models? SigLIP models represent an advanced variant of CLIP models in the OpenCLIP framework , featuring sigmoid-based contrastive learning and several architectural modifications. These models are available in multiple resolution variants and can be used with the standard OpenCLIP interfaces for training and inference. Which siglip model is best for training? The training configuration will need to specify the sigmoid-based loss instead of the standard softmax-based CLIP loss. Higher-resolution SigLIP models (384×384, 512×512) generally provide better performance but require more computational resources for both training and inference. Where can I find an open source implementation of clip? GitHub - mlfoundations/open_clip: An open source implementation of CLIP. Move src/clip -> src/open_clip, create setup.py + MANIFEST.in and tes… [Paper] [Citations] [Clip Colab] [Coca Colab] Welcome to an open source implementation of OpenAI's CLIP (Contrastive Language-Image Pre-training). Does siglip outperform clip? A final cross-device sum brings everything together. SigLIP outperforms CLIP at small batch sizes (e.g., 4–8k), but both reach saturation at 32k batch size despite claims that larger batches improve performance. LiT: SigLIP results, trained for 9B seen examples. Aug 6, 2025 · OpenCLIP [Paper] [Citations] [Clip Colab] [Coca Colab] Welcome to an open source implementation of OpenAI's CLIP ( Contrastive Language-Image Pre-training). Using this codebase, we have trained several models on a variety of data sources and compute budgets, ranging from small-scale experiments to larger runs including models trained on datasets such as LAION-400M, LAION-2B and DataComp-1B ..."}
+{"idx": 2, "title": "SigLIP vs. CLIP: The Sigmoid Advantage | by heping_LU | Medium", "date": "", "ddg_snippet": "Sep 25, 2024 · Contrastive pre-training, using weakly supervised image-text pairs, has become the leading method for developing general computer vision models . This involves learning aligned representations for ...", "subpage_snippet": "", "source": "medium.com", "link": "https://medium.com/@jiangmen28/siglip-vs-clip-the-sigmoid-advantage-457f1cb872ab", "content": "Sep 25, 2024 · Contrastive pre-training, using weakly supervised image-text pairs, has become the leading method for developing general computer vision models . This involves learning aligned representations for ..."}
+{"idx": 3, "title": "CLIP to SigLIP: Vision-Language Models with Contrastive Learning", "date": "", "ddg_snippet": "Aug 2, 2024 · Understand the changes from CLIP to SigLIP in vision-language models , highlighting the advantages of contrastive learning and how they enhances efficiency", "subpage_snippet": "", "source": "blog.ritwikraha.dev", "link": "https://blog.ritwikraha.dev/choosing-between-siglip-and-clip-for-language-image-pretraining", "content": "Aug 2, 2024 · Understand the changes from CLIP to SigLIP in vision-language models , highlighting the advantages of contrastive learning and how they enhances efficiency"}
+{"idx": 4, "title": "Inference_with_ (multilingual)_SigLIP,_a_better_CLIP_model ...", "date": "", "ddg_snippet": "The sigmoid loss simultaneously allows further scaling up the batch size during pre-training (like 32k image-text pairs), while also performing better at smaller batch sizes. The model outperforms CLIP on both zero-shot image classification and image-text retrieval as shown below.", "subpage_snippet": "", "source": "colab.research.google.com", "link": "https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/SigLIP/Inference_with_(multilingual)_SigLIP,_a_better_CLIP_model.ipynb", "content": "The sigmoid loss simultaneously allows further scaling up the batch size during pre-training (like 32k image-text pairs), while also performing better at smaller batch sizes. The model outperforms CLIP on both zero-shot image classification and image-text retrieval as shown below."}
+{"idx": 5, "title": "GitHub - ramanakshay/clip: CLIP & SigLIP model training from ...", "date": "", "ddg_snippet": "Contrastive Language-Image Pre-training (CLIP) is a technique for training a pair of neural network models , one for image understanding and one for text understanding, using a contrastive objective. This project contains code to train CLIP on the MS-COCO Captions dataset. It also includes an implementation of SigLIP, which uses a sigmoid loss as the training objective.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/ramanakshay/clip", "content": "Contrastive Language-Image Pre-training (CLIP) is a technique for training a pair of neural network models , one for image understanding and one for text understanding, using a contrastive objective. This project contains code to train CLIP on the MS-COCO Captions dataset. It also includes an implementation of SigLIP, which uses a sigmoid loss as the training objective."}
+{"idx": 6, "title": "open-clip-torch · PyPI", "date": "", "ddg_snippet": "Aug 6, 2025 · OpenCLIP [Paper] [Citations] [Clip Colab] [Coca Colab] Welcome to an open source implementation of OpenAI's CLIP ( Contrastive Language-Image Pre-training). Using this codebase, we have trained several models on a variety of data sources and compute budgets, ranging from small-scale experiments to larger runs including models trained on datasets such as LAION-400M, LAION-2B and DataComp-1B ...", "subpage_snippet": "", "source": "pypi.org", "link": "https://pypi.org/project/open-clip-torch/", "content": "Aug 6, 2025 · OpenCLIP [Paper] [Citations] [Clip Colab] [Coca Colab] Welcome to an open source implementation of OpenAI's CLIP ( Contrastive Language-Image Pre-training). Using this codebase, we have trained several models on a variety of data sources and compute budgets, ranging from small-scale experiments to larger runs including models trained on datasets such as LAION-400M, LAION-2B and DataComp-1B ..."}
+{"idx": 7, "title": "lesswrong.com/posts/r6gpBgs98gnArCEty/how-to-think-with-images", "date": "", "ddg_snippet": "Contrastive Language-Image Pretraining : While generative models were evolving, another thread of research tackled the understanding side of vision ...", "subpage_snippet": "", "source": "www.lesswrong.com", "link": "https://www.lesswrong.com/posts/r6gpBgs98gnArCEty/how-to-think-with-images", "content": "Contrastive Language-Image Pretraining : While generative models were evolving, another thread of research tackled the understanding side of vision ..."}
+{"idx": 8, "title": "multimodal AI | LearnOpenCV", "date": "", "ddg_snippet": "... Contrastive Learning Decoder-Based Pretraining deep learning Dense Features AI Equitable AI FSDP Gemma Tokenizer Google DeepMind Image Captioning AI ...", "subpage_snippet": "", "source": "learnopencv.com", "link": "https://learnopencv.com/tag/multimodal-ai/", "content": "... Contrastive Learning Decoder-Based Pretraining deep learning Dense Features AI Equitable AI FSDP Gemma Tokenizer Google DeepMind Image Captioning AI ..."}
+{"idx": 9, "title": "TokLIP: Marry Visual Tokens to CLIP for Multimodal", "date": "", "ddg_snippet": "Exploring unified modeling of multimodal comprehension and generation in the framework of large language models has gained significant attention.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2505.05422v1", "content": "Exploring unified modeling of multimodal comprehension and generation in the framework of large language models has gained significant attention."}
diff --git a/data/sampled_jsons/OpenReview_0cEZyhHEks_Taming_Knowledge_Conflicts_Table_3.jsonl b/data/sampled_jsons/OpenReview_0cEZyhHEks_Taming_Knowledge_Conflicts_Table_3.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..5dd04294feeb311f84a02bfe7905295915e8fb29
--- /dev/null
+++ b/data/sampled_jsons/OpenReview_0cEZyhHEks_Taming_Knowledge_Conflicts_Table_3.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Taming Knowledge Conflicts in Language Models - OpenReview", "date": "", "ddg_snippet": "Abstract Language Models (LMs) often encounter knowl-edge conflicts when parametric memory con-tradicts contextual knowledge . Previous works attribute this conflict to the interplay between \"memory heads\" and \"context heads\", attention heads assumed to promote either memory or con-text exclusively. In this study, we go beyond this fundamental assumption by uncovering a criti-cal ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=0cEZyhHEks", "content": "Abstract Language Models (LMs) often encounter knowl-edge conflicts when parametric memory con-tradicts contextual knowledge . Previous works attribute this conflict to the interplay between \"memory heads\" and \"context heads\", attention heads assumed to promote either memory or con-text exclusively. In this study, we go beyond this fundamental assumption by uncovering a criti-cal ..."}
+{"idx": 1, "title": "Taming Knowledge Conflicts in Language Models - GitHub", "date": "", "ddg_snippet": "This repository contains the code and data of the ICML 25 Spotlight Paper Taming Knowledge Conflicts in Language Models. The code is now still being updated.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/GaotangLi/JUICE", "content": "This repository contains the code and data of the ICML 25 Spotlight Paper Taming Knowledge Conflicts in Language Models. The code is now still being updated."}
+{"idx": 2, "title": "Knowledge Conflicts for LLMs: A Survey - ACL Anthology", "date": "", "ddg_snippet": "Abstract This survey provides an in-depth analysis of knowledge conflicts for large language models (LLMs), highlighting the complex challenges they encounter when blending contextual and parametric knowledge . Our focus is on three categories of knowledge conflicts : context-memory, inter-context, and intra-memory conflict .", "subpage_snippet": "", "source": "aclanthology.org", "link": "https://aclanthology.org/2024.emnlp-main.486/", "content": "Abstract This survey provides an in-depth analysis of knowledge conflicts for large language models (LLMs), highlighting the complex challenges they encounter when blending contextual and parametric knowledge . Our focus is on three categories of knowledge conflicts : context-memory, inter-context, and intra-memory conflict ."}
+{"idx": 3, "title": "Resolving Knowledge Conflicts in Large Language Models - OpenReview", "date": "", "ddg_snippet": "We posit that LLMs should 1) identify knowledge conflicts , 2) pinpoint conflicting information segments, and 3 ) provide distinct answers or viewpoints in conflicting scenarios. To this end, we introduce an evaluation framework for simulating contextual knowledge conflicts and quantitatively evaluating to what extent LLMs achieve these goals.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=ptvV5HGTNN", "content": "We posit that LLMs should 1) identify knowledge conflicts , 2) pinpoint conflicting information segments, and 3 ) provide distinct answers or viewpoints in conflicting scenarios. To this end, we introduce an evaluation framework for simulating contextual knowledge conflicts and quantitatively evaluating to what extent LLMs achieve these goals."}
+{"idx": 4, "title": "Knowledge Conflicts for LLMs: A Survey - OpenReview", "date": "", "ddg_snippet": "This survey provides an in-depth analysis of knowledge conflicts for large language models (LLMs), highlighting the complex challenges they encounter when blending contextual and parametric knowledge . Our focus is on three categories of knowledge conflicts : context-memory, inter-context, and intra-memory conflict . These conflicts can significantly impact the trustworthiness and performance of ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=8l1Vm1p1Jv", "content": "This survey provides an in-depth analysis of knowledge conflicts for large language models (LLMs), highlighting the complex challenges they encounter when blending contextual and parametric knowledge . Our focus is on three categories of knowledge conflicts : context-memory, inter-context, and intra-memory conflict . These conflicts can significantly impact the trustworthiness and performance of ..."}
+{"idx": 5, "title": "ConflictBank: A Benchmark for Evaluating the Influence of Knowledge ...", "date": "", "ddg_snippet": "However, a thorough assessment of knowledge conflict in LLMs is still missing. Motivated by this research gap, we present ConflictBank, the first comprehensive benchmark developed to systematically evaluate knowledge conflicts from three aspects: (i) conflicts encountered in retrieved knowledge , (ii) conflicts within the models' encoded ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2408.12076", "content": "However, a thorough assessment of knowledge conflict in LLMs is still missing. Motivated by this research gap, we present ConflictBank, the first comprehensive benchmark developed to systematically evaluate knowledge conflicts from three aspects: (i) conflicts encountered in retrieved knowledge , (ii) conflicts within the models' encoded ..."}
+{"idx": 6, "title": "ICML Taming Knowledge Conflicts in Language Models", "date": "", "ddg_snippet": "Poster in Workshop: Actionable Interpretability Taming Knowledge Conflicts in Language Models Gaotang Li · Yuzhong Chen · Hanghang Tong [ Abstract ] [ Project Page ] [ OpenReview ] Sat 19 Jul 10:40 a.m. PDT — 11:40 a.m. PDT", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/49596", "content": "Poster in Workshop: Actionable Interpretability Taming Knowledge Conflicts in Language Models Gaotang Li · Yuzhong Chen · Hanghang Tong [ Abstract ] [ Project Page ] [ OpenReview ] Sat 19 Jul 10:40 a.m. PDT — 11:40 a.m. PDT"}
+{"idx": 7, "title": "Taming Knowledge Conflicts in Language Models", "date": "", "ddg_snippet": "Language Models (LMs) often encounter knowledge conflicts when parametric memory contradicts contextual knowledge . Previous works attribute this conflict to the interplay between \"memory heads\" and \"context heads\", attention heads assumed to promote either memory or context exclusively. In this study, we go beyond this fundamental assumption by uncovering a critical phenomenon we term ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2503.10996v2", "content": "Language Models (LMs) often encounter knowledge conflicts when parametric memory contradicts contextual knowledge . Previous works attribute this conflict to the interplay between \"memory heads\" and \"context heads\", attention heads assumed to promote either memory or context exclusively. In this study, we go beyond this fundamental assumption by uncovering a critical phenomenon we term ..."}
+{"idx": 8, "title": "Forum | OpenReview", "date": "", "ddg_snippet": "Promoting openness in scientific communication and the peer-review process", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=0cEZyhHEks", "content": "Promoting openness in scientific communication and the peer-review process"}
+{"idx": 9, "title": "Cutting Off the Head Ends the Conflict: A Mechanism for Interpreting ...", "date": "", "ddg_snippet": "However, internal memory and external context inevitably clash, leading to knowledge conflicts within LMs. In this paper, we aim to interpret the mechanism of knowledge conflicts through the lens of information flow, and then mitigate conflicts by precise interventions at the pivotal point.", "subpage_snippet": "", "source": "aclanthology.org", "link": "https://aclanthology.org/2024.findings-acl.70/", "content": "However, internal memory and external context inevitably clash, leading to knowledge conflicts within LMs. In this paper, we aim to interpret the mechanism of knowledge conflicts through the lens of information flow, and then mitigate conflicts by precise interventions at the pivotal point."}
diff --git a/data/sampled_jsons/RAGGED_Jennifer_Hsia_Afreen_Shaikh_retrieval_augmented_generation_noise_sensitivity_reader_models_year_2024.jsonl b/data/sampled_jsons/RAGGED_Jennifer_Hsia_Afreen_Shaikh_retrieval_augmented_generation_noise_sensitivity_reader_models_year_2024.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..b4bb640215636458fb70756a63db4c45dbf5ecce
--- /dev/null
+++ b/data/sampled_jsons/RAGGED_Jennifer_Hsia_Afreen_Shaikh_retrieval_augmented_generation_noise_sensitivity_reader_models_year_2024.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "RAGGED: Towards Informed Design of Retrieval ...", "date": "", "ddg_snippet": "by J Hsia · Cited by 23 — It investigates how various components—such as different retrievers (BM25, ColBERT, Contriever) and reader models (FLAN, LLaMa, GPT, Claude)—impact RAG ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=KDXj60FpJr", "content": "by J Hsia · Cited by 23 — It investigates how various components—such as different retrievers (BM25, ColBERT, Contriever) and reader models (FLAN, LLaMa, GPT, Claude)—impact RAG ..."}
+{"idx": 1, "title": "RAGGED: Towards Informed Design of Retrieval ...", "date": "", "ddg_snippet": "RAGGED: Towards Informed Design of Retrieval Augmented Generation Systems . Jennifer Hsia · Afreen Shaikh · Zhiruo Wang · Graham Neubig.", "subpage_snippet": "", "source": "neurips.cc", "link": "https://neurips.cc/virtual/2024/104956", "content": "RAGGED: Towards Informed Design of Retrieval Augmented Generation Systems . Jennifer Hsia · Afreen Shaikh · Zhiruo Wang · Graham Neubig."}
+{"idx": 2, "title": "RAGGED: Towards Informed Design of Retrieval- ...", "date": "", "ddg_snippet": "by J Hsia · Cited by 23 — Solution: The RAGGED framework, a systematic tool for optimizing RAG configurations. Key Takeaways. 1. Suboptimal RAG can be worse than no-context. 2. Reader ...", "subpage_snippet": "", "source": "adaptive-foundation-models.org", "link": "https://adaptive-foundation-models.org/posters/RAGGED_AFM_2024_Poster.pdf", "content": "by J Hsia · Cited by 23 — Solution: The RAGGED framework, a systematic tool for optimizing RAG configurations. Key Takeaways. 1. Suboptimal RAG can be worse than no-context. 2. Reader ..."}
+{"idx": 3, "title": "neulab/ragged: Retrieval Augmented Generation ...", "date": "", "ddg_snippet": "In this work, we introduce RAGGED , a framework for systematically evaluating RAG systems across diverse retriever-reader configurations, retrieval depths, and ...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/neulab/ragged", "content": "In this work, we introduce RAGGED , a framework for systematically evaluating RAG systems across diverse retriever-reader configurations, retrieval depths, and ..."}
+{"idx": 4, "title": "RAGGED: Towards Informed Design of Retrieval ...", "date": "", "ddg_snippet": "by J Hsia · Cited by 23 — Figure 1: Example insight from using RAGGED: LLAMA and CLAUDE models are more sensitive to noise in context, while FLAN and GPT models are more robust to noise ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=SX14yxTTRB", "content": "by J Hsia · Cited by 23 — Figure 1: Example insight from using RAGGED: LLAMA and CLAUDE models are more sensitive to noise in context, while FLAN and GPT models are more robust to noise ..."}
+{"idx": 5, "title": "RAGGED: Optimized RAG System Design", "date": "", "ddg_snippet": "The paper introduces the RAGGED framework that systematically optimizes retriever and reader configurations for DBQA tasks. It demonstrates that encoder-decoder ...", "subpage_snippet": "", "source": "www.emergentmind.com", "link": "https://www.emergentmind.com/papers/2403.09040", "content": "The paper introduces the RAGGED framework that systematically optimizes retriever and reader configurations for DBQA tasks. It demonstrates that encoder-decoder ..."}
+{"idx": 6, "title": "Towards Informed Design of Scalable and Stable RAG ...", "date": "", "ddg_snippet": "14 Mar 2024 — Retrieval - augmented generation ( RAG ) enhances language models by integrating external knowledge, but its effectiveness is highly dependent on ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2403.09040", "content": "14 Mar 2024 — Retrieval - augmented generation ( RAG ) enhances language models by integrating external knowledge, but its effectiveness is highly dependent on ..."}
+{"idx": 7, "title": "Towards Informed Design of Scalable and Stable RAG ...", "date": "", "ddg_snippet": "To address these challenges, we introduce RAGGED (Retrieval-Augmented Generation Generalized Evaluation Device), a framework for systematically evaluating RAG ...", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/poster/46460", "content": "To address these challenges, we introduce RAGGED (Retrieval-Augmented Generation Generalized Evaluation Device), a framework for systematically evaluating RAG ..."}
+{"idx": 8, "title": "Towards Informed Design of Scalable and Stable RAG ...", "date": "", "ddg_snippet": "We introduce RAGGED , a framework for systematically evaluating RAG systems across diverse retriever-reader configurations, retrieval depths, and datasets.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2403.09040v3", "content": "We introduce RAGGED , a framework for systematically evaluating RAG systems across diverse retriever-reader configurations, retrieval depths, and datasets."}
+{"idx": 9, "title": "Towards Informed Design of Scalable and Stable RAG Systems", "date": "", "ddg_snippet": "In this work, we introduce RAGGED , a framework for systematically evaluating RAG systems across diverse retriever-reader configurations, retrieval depths, and ...", "subpage_snippet": "", "source": "chatpaper.com", "link": "https://chatpaper.com/fr/chatpaper/paper/165564", "content": "In this work, we introduce RAGGED , a framework for systematically evaluating RAG systems across diverse retriever-reader configurations, retrieval depths, and ..."}
diff --git a/data/sampled_jsons/RAGGED_RAG_Stability_Score_RSS_formula_Jennifer_Hsia.jsonl b/data/sampled_jsons/RAGGED_RAG_Stability_Score_RSS_formula_Jennifer_Hsia.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..d7977dbbad811454db337b3dd0189b208027e351
--- /dev/null
+++ b/data/sampled_jsons/RAGGED_RAG_Stability_Score_RSS_formula_Jennifer_Hsia.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "[2403.09040] RAGGED : Towards Informed Design of Scalable and...", "date": "", "ddg_snippet": "In this work, we introduce RAGGED , a framework for systematically evaluating RAG systems across diverse retriever-reader configurations, retrieval depths, and datasets. Our analysis reveals that reader robustness to noise is the key determinant of RAG stability and scalability.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2403.09040", "content": "In this work, we introduce RAGGED , a framework for systematically evaluating RAG systems across diverse retriever-reader configurations, retrieval depths, and datasets. Our analysis reveals that reader robustness to noise is the key determinant of RAG stability and scalability."}
+{"idx": 1, "title": "RAG (Retrieval Augmented Generation) — простое и понятное... / Хабр", "date": "", "ddg_snippet": "RAG (Retrieval Augmented Generation) — это метод работы с большими языковыми моделями, когда пользователь пишет свой вопросы, а вы программно к этому вопросу «подмешиваете» дополнительную информацию из каких‑то внешних источников и подаете все...", "subpage_snippet": "", "source": "habr.com", "link": "https://habr.com/ru/articles/779526/", "content": "RAG (Retrieval Augmented Generation) — это метод работы с большими языковыми моделями, когда пользователь пишет свой вопросы, а вы программно к этому вопросу «подмешиваете» дополнительную информацию из каких‑то внешних источников и подаете все..."}
+{"idx": 2, "title": "Top Research Papers on RAG", "date": "", "ddg_snippet": "RAGGED , a framework for analyzing RAG configurations across various DBQA tasks, discovers distinct LM behaviors in response to varying context quantities, context qualities, and retrievers and provides a deeper analysis of these differences.", "subpage_snippet": "", "source": "paperguide.ai", "link": "https://paperguide.ai/papers/top/research-papers-rag/", "content": "RAGGED , a framework for analyzing RAG configurations across various DBQA tasks, discovers distinct LM behaviors in response to varying context quantities, context qualities, and retrievers and provides a deeper analysis of these differences."}
+{"idx": 3, "title": "Evaluation framework for your Retrieval Augmented Generation ( RAG )...", "date": "", "ddg_snippet": "Evaluate a simple RAG .JSON Parsing Issue: The model's output is not JSON-parsable. ragas requires models to output JSON-compatible responses because all prompts are structured using Pydantic. This ensures efficient parsing of LLM outputs. Non-Ideal Cases for Scoring ...", "subpage_snippet": "", "source": "docs.ragas.io", "link": "https://docs.ragas.io/", "content": "Evaluate a simple RAG .JSON Parsing Issue: The model's output is not JSON-parsable. ragas requires models to output JSON-compatible responses because all prompts are structured using Pydantic. This ensures efficient parsing of LLM outputs. Non-Ideal Cases for Scoring ..."}
+{"idx": 4, "title": "Graham Neubig's Publications", "date": "", "ddg_snippet": "Jennifer Hsia , Afreen Shaikh, Zora Zhiruo Wang, Graham Neubig. RAGGED : Towards Informed Design of Scalable and Stable RAG Systems (BibTex, Code/Data) International Conference on Machine Learning (ICML).", "subpage_snippet": "", "source": "phontron.com", "link": "https://phontron.com/publications.php", "content": "Jennifer Hsia , Afreen Shaikh, Zora Zhiruo Wang, Graham Neubig. RAGGED : Towards Informed Design of Scalable and Stable RAG Systems (BibTex, Code/Data) International Conference on Machine Learning (ICML)."}
+{"idx": 5, "title": "The Love Song of J. Alfred Prufrock | The Poetry Foundation", "date": "", "ddg_snippet": "And I have known the eyes already, known them all—The eyes that fix you in a formulated phraseAnd when I am formulated , sprawling on a pin", "subpage_snippet": "", "source": "www.poetryfoundation.org", "link": "https://www.poetryfoundation.org/poetrymagazine/poems/44212/the-love-song-of-j-alfred-prufrock", "content": "And I have known the eyes already, known them all—The eyes that fix you in a formulated phraseAnd when I am formulated , sprawling on a pin"}
+{"idx": 6, "title": "Carnegie Mellon University at ICML 2025 – Machine Learning Blog", "date": "", "ddg_snippet": "RAGGED : Towards Informed Design of Scalable and Stable RAG Systems. Authors: Jennifer Hsia , Afreen Shaikh, Zhiruo Wang, Graham Neubig.", "subpage_snippet": "", "source": "blog.ml.cmu.edu", "link": "https://blog.ml.cmu.edu/2025/07/08/carnegie-mellon-university-at-icml-2025/", "content": "RAGGED : Towards Informed Design of Scalable and Stable RAG Systems. Authors: Jennifer Hsia , Afreen Shaikh, Zhiruo Wang, Graham Neubig."}
+{"idx": 7, "title": "Delete USA, by Hua Bin - The Unz Review", "date": "", "ddg_snippet": "The Trump administration and a rag tag collection of US tech/crypto oligarchs has conned the entire US economy into thinking there’s a “race to AI dominance.” Even worse than that, they’ve convinced America that we must win it.", "subpage_snippet": "", "source": "www.unz.com", "link": "https://www.unz.com/bhua/delete-usa/", "content": "The Trump administration and a rag tag collection of US tech/crypto oligarchs has conned the entire US economy into thinking there’s a “race to AI dominance.” Even worse than that, they’ve convinced America that we must win it."}
+{"idx": 8, "title": "Farcical scenes in the Championship as Blackburn... | Daily Mail Online", "date": "", "ddg_snippet": "Formula One's first female president hopeful, 28, opens up on taking on men - and fights back at questions over her Instagram bikini photos.", "subpage_snippet": "", "source": "www.dailymail.co.uk", "link": "https://www.dailymail.co.uk/sport/football/article-15117763/Championship-Blackburn-vs-Ipswich-called-off.html", "content": "Formula One's first female president hopeful, 28, opens up on taking on men - and fights back at questions over her Instagram bikini photos."}
+{"idx": 9, "title": "Henry Halfhead - Review - Gamer Social Club", "date": "", "ddg_snippet": "Gamer Social Club Review Score Policy. Henry Halfhead is available now on PC(Steam, EGS), PS5, and Nintendo Switch. Formula Legends – Accurate Driver Achievement Guide. Dan “Danno” Jackson.", "subpage_snippet": "", "source": "gamersocialclub.ca", "link": "https://gamersocialclub.ca/2025/09/18/henry-halfhead-review/", "content": "Gamer Social Club Review Score Policy. Henry Halfhead is available now on PC(Steam, EGS), PS5, and Nintendo Switch. Formula Legends – Accurate Driver Achievement Guide. Dan “Danno” Jackson."}
diff --git a/data/sampled_jsons/RAGGED_paper_Towards_Informed_Design_of_Scalable_and_Stable_RAG_Systems_Section_3.1_year_2023.jsonl b/data/sampled_jsons/RAGGED_paper_Towards_Informed_Design_of_Scalable_and_Stable_RAG_Systems_Section_3.1_year_2023.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..36bbb8a52fd5407ce97c8834662a9697719b343d
--- /dev/null
+++ b/data/sampled_jsons/RAGGED_paper_Towards_Informed_Design_of_Scalable_and_Stable_RAG_Systems_Section_3.1_year_2023.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Engineering RAG Systems for Real-World Applications: Design,", "date": "", "ddg_snippet": "... of this paper are as follows: end-to-end development and deployment of RAG systems for multilingual, domain-specific applications; user-centred ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2506.20869v2", "content": "... of this paper are as follows: end-to-end development and deployment of RAG systems for multilingual, domain-specific applications; user-centred ..."}
+{"idx": 1, "title": "Domain-Aware RAG: MoL-Enhanced RL for Efficient Training and", "date": "", "ddg_snippet": "MoLER systematically bridges the gap between conventional enhancement techniques and the evolving demands of RAG systems , demonstrating significant ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2509.06650v1", "content": "MoLER systematically bridges the gap between conventional enhancement techniques and the evolving demands of RAG systems , demonstrating significant ..."}
+{"idx": 2, "title": "RAG-R1 : Incentivize the Search and Reasoning Capabilities of", "date": "", "ddg_snippet": "In this paper , we propose RAG -R1, a novel training framework designed to enable LLMs to adaptively leverage internal and external knowledge during ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2507.02962v4", "content": "In this paper , we propose RAG -R1, a novel training framework designed to enable LLMs to adaptively leverage internal and external knowledge during ..."}
+{"idx": 3, "title": "How to Build a RAG System That Actually Works!", "date": "", "ddg_snippet": "An effective retrieval system is the backbone of RAG , ensuring that the model has access to accurate, relevant, and contextually rich information.", "subpage_snippet": "", "source": "www.analyticsvidhya.com", "link": "https://www.analyticsvidhya.com/blog/2025/03/why-rag-systems-fail-and-how-to-fix-them/", "content": "An effective retrieval system is the backbone of RAG , ensuring that the model has access to accurate, relevant, and contextually rich information."}
+{"idx": 4, "title": "Building Contextual RAG Systems with Hybrid Search &", "date": "", "ddg_snippet": "... article we will focus particularly on solving the limitations of naive RAG systems in terms of adding contextual information to document chunks and ...", "subpage_snippet": "", "source": "www.analyticsvidhya.com", "link": "https://www.analyticsvidhya.com/blog/2024/12/contextual-rag-systems-with-hybrid-search-and-reranking/", "content": "... article we will focus particularly on solving the limitations of naive RAG systems in terms of adding contextual information to document chunks and ..."}
+{"idx": 5, "title": "RAG+: Enhancing Retrieval-Augmented Generation with", "date": "", "ddg_snippet": "... and open-domain question answering, RAG frequently underperforms on domain-specific reasoning tasks [ Section 1 ], where solving complex problems ...", "subpage_snippet": "", "source": "keyurramoliya.com", "link": "https://keyurramoliya.com/posts/Rag-Plus/", "content": "... and open-domain question answering, RAG frequently underperforms on domain-specific reasoning tasks [ Section 1 ], where solving complex problems ..."}
+{"idx": 6, "title": "Benchmarking RAG Systems: Making AI Answers Reliable, Fast, and", "date": "", "ddg_snippet": "... handled by the RAG system ? The knowledge base may increase from thousands to millions of papers as businesses grow, and traffic may increase as well.", "subpage_snippet": "", "source": "www.walturn.com", "link": "https://www.walturn.com/insights/benchmarking-rag-systems-making-ai-answers-reliable-fast-and-useful", "content": "... handled by the RAG system ? The knowledge base may increase from thousands to millions of papers as businesses grow, and traffic may increase as well."}
+{"idx": 7, "title": "How RAG Systems Improve Public Sector Management -", "date": "", "ddg_snippet": "In a RAG system , the retriever is the first step of the RAG pipeline and is responsible for locating and obtaining relevant information from a vast ...", "subpage_snippet": "", "source": "www.datategy.net", "link": "https://www.datategy.net/2025/04/07/how-rag-systems-improve-public-sector-management/", "content": "In a RAG system , the retriever is the first step of the RAG pipeline and is responsible for locating and obtaining relevant information from a vast ..."}
+{"idx": 8, "title": "12 RAG Pain Points and Proposed Solutions | Towards Data Science", "date": "", "ddg_snippet": "... help in situations where the system might otherwise provide a plausible but incorrect answer due to the lack of information in the knowledge base.", "subpage_snippet": "", "source": "towardsdatascience.com", "link": "https://towardsdatascience.com/12-rag-pain-points-and-proposed-solutions-43709939a28c/", "content": "... help in situations where the system might otherwise provide a plausible but incorrect answer due to the lack of information in the knowledge base."}
+{"idx": 9, "title": "Solving AI Foundational Model Latency with Telco Infrastructure", "date": "", "ddg_snippet": "The remainder of this paper is organized as follows: Section 2 surveys foundational model architectures, inference bottlenecks, and related caching ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2504.03708v1", "content": "The remainder of this paper is organized as follows: Section 2 surveys foundational model architectures, inference bottlenecks, and related caching ..."}
diff --git a/data/sampled_jsons/RepE_acronym_machine_learning.jsonl b/data/sampled_jsons/RepE_acronym_machine_learning.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..2f204a9f85495546601b6c4628c6628e407d9180
--- /dev/null
+++ b/data/sampled_jsons/RepE_acronym_machine_learning.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "REPE vs REI", "date": "", "ddg_snippet": "6 Aug 2020 — REPE firms focus on raising capital and financial analysis , while REI firms focus on specific deals and hands-on property experience.", "subpage_snippet": "", "source": "www.wallstreetoasis.com", "link": "https://www.wallstreetoasis.com/forum/real-estate/repe-vs-rei", "content": "6 Aug 2020 — REPE firms focus on raising capital and financial analysis , while REI firms focus on specific deals and hands-on property experience."}
+{"idx": 1, "title": "Real Estate Private Equity (REPE)", "date": "", "ddg_snippet": "Real Estate Private Equity (REPE ) firms raise capital to acquire, develop, improve, and operate real estate, focusing on commercial and multifamily properties.", "subpage_snippet": "", "source": "corporatefinanceinstitute.com", "link": "https://corporatefinanceinstitute.com/resources/commercial-real-estate/real-estate-private-equity/", "content": "Real Estate Private Equity (REPE ) firms raise capital to acquire, develop, improve, and operate real estate, focusing on commercial and multifamily properties."}
+{"idx": 2, "title": "Difference between REPE, GP, LP, debt fund?", "date": "", "ddg_snippet": "28 Jul 2020 — REPE - any private equity fund that invests in real estate assets. This could be taking the GP, LP, or senior lending (debt fund) position. GP - ...", "subpage_snippet": "", "source": "www.wallstreetoasis.com", "link": "https://www.wallstreetoasis.com/forum/real-estate/difference-between-repe-gp-lp-debt-fund", "content": "28 Jul 2020 — REPE - any private equity fund that invests in real estate assets. This could be taking the GP, LP, or senior lending (debt fund) position. GP - ..."}
+{"idx": 3, "title": "Importance-Aware Learning for Neural Headline Editing", "date": "", "ddg_snippet": "by Q Wu · 2020 · Cited by 19 — We use the acronym . PAS, which stands for Pre-training, Adaptation, and SIA, to name our final model. Figure 2 describes the overall pipeline of PAS. Pre- ... 8 pages", "subpage_snippet": "", "source": "cdn.aaai.org", "link": "https://cdn.aaai.org/ojs/6467/6467-13-9692-1-10-20200517.pdf", "content": "by Q Wu · 2020 · Cited by 19 — We use the acronym . PAS, which stands for Pre-training, Adaptation, and SIA, to name our final model. Figure 2 describes the overall pipeline of PAS. Pre- ... 8 pages"}
+{"idx": 4, "title": "Radar pulse repetition interval modulation recognition with ...", "date": "", "ddg_snippet": "by HC Feng · 2022 · Cited by 27 — In this paper, we design a multiscale convolution block with a vectorized embedding and squeeze-and-excitation mechanism for feature extraction.", "subpage_snippet": "", "source": "www.sciencedirect.com", "link": "https://www.sciencedirect.com/science/article/pii/S1051200422001798", "content": "by HC Feng · 2022 · Cited by 27 — In this paper, we design a multiscale convolution block with a vectorized embedding and squeeze-and-excitation mechanism for feature extraction."}
+{"idx": 5, "title": "Collective Information Extraction with Relational Markov ...", "date": "", "ddg_snippet": "by R Bunescu · 2004 · Cited by 177 — This allows for \"collective informa- tion extraction\" that exploits the mutual in- fluence between possible extractions. Experi- ments on learning to extract ...", "subpage_snippet": "", "source": "aclanthology.org", "link": "https://aclanthology.org/P04-1056.pdf", "content": "by R Bunescu · 2004 · Cited by 177 — This allows for \"collective informa- tion extraction\" that exploits the mutual in- fluence between possible extractions. Experi- ments on learning to extract ..."}
+{"idx": 6, "title": "AI System-to-Model Innovation | CSET", "date": "", "ddg_snippet": "by J Schiestle · 2025 — Calling the approach representation engineering (RepE), the researchers presented unsupervised methods for “reading” and “control.”43 Reading finds the ...", "subpage_snippet": "", "source": "cset.georgetown.edu", "link": "https://cset.georgetown.edu/wp-content/uploads/CSET-AI-System-to-Model-Innovation.pdf", "content": "by J Schiestle · 2025 — Calling the approach representation engineering (RepE), the researchers presented unsupervised methods for “reading” and “control.”43 Reading finds the ..."}
+{"idx": 7, "title": "Balancing Stylization and Truth via Disentangled ...", "date": "", "ddg_snippet": "by C Shen · 2025 · Cited by 1 — Abstract. Generating stylized large language model (LLM) responses via rep- resentation editing is a promising way for fine-grained output.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2508.04530?", "content": "by C Shen · 2025 · Cited by 1 — Abstract. Generating stylized large language model (LLM) responses via rep- resentation editing is a promising way for fine-grained output."}
+{"idx": 8, "title": "Collective Information Extraction with Relational Markov ...", "date": "", "ddg_snippet": "by R Bunescu · Cited by 177 — Relational Markov Networks (RMNs) are a generalization of CRFs that represent dependencies between extractions, enabling collective information extraction. ...", "subpage_snippet": "", "source": "www.cs.utexas.edu", "link": "https://www.cs.utexas.edu/~ml/papers/cie-acl-04.pdf", "content": "by R Bunescu · Cited by 177 — Relational Markov Networks (RMNs) are a generalization of CRFs that represent dependencies between extractions, enabling collective information extraction. ..."}
+{"idx": 9, "title": "SUGAR: Pre-training 3D Visual Representations for Robotics", "date": "", "ddg_snippet": "by S Chen · 2024 · Cited by 23 — SUGAR is a 3D pre-training framework for robotics that learns semantic, geometric, and affordance properties of objects in 3D point clouds.", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content/CVPR2024/papers/Chen_SUGAR_Pre-training_3D_Visual_Representations_for_Robotics_CVPR_2024_paper.pdf", "content": "by S Chen · 2024 · Cited by 23 — SUGAR is a 3D pre-training framework for robotics that learns semantic, geometric, and affordance properties of objects in 3D point clouds."}
diff --git a/data/sampled_jsons/S18_characters_dataset_Machine_Learning_meets_Algebraic_Combinatorics_training_examples_Appendix_B.1.jsonl b/data/sampled_jsons/S18_characters_dataset_Machine_Learning_meets_Algebraic_Combinatorics_training_examples_Appendix_B.1.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..901448581d523ba0f0aec14f28864b865db0c351
--- /dev/null
+++ b/data/sampled_jsons/S18_characters_dataset_Machine_Learning_meets_Algebraic_Combinatorics_training_examples_Appendix_B.1.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Machine Learning meets Algebraic Combinatorics", "date": "", "ddg_snippet": "by H Chau · 2025 · Cited by 3 — We describe all nine datasets , the different ways machine learning models can be applied to them (e.g., training with narrow models followed by ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2503.06366", "content": "by H Chau · 2025 · Cited by 3 — We describe all nine datasets , the different ways machine learning models can be applied to them (e.g., training with narrow models followed by ..."}
+{"idx": 1, "title": "Machine Learning Meets Algebraic Combinatorics", "date": "", "ddg_snippet": "See Appendix B . 1 . • Does the dataset contain all possible instances or is it a sample (not necessarily random) of instances from a larger set? It contains ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/attachment?id=tlniJJFUW2&name=pdf", "content": "See Appendix B . 1 . • Does the dataset contain all possible instances or is it a sample (not necessarily random) of instances from a larger set? It contains ..."}
+{"idx": 2, "title": "Appendix D", "date": "", "ddg_snippet": "While machine learning techniques are enabling breakthroughs in assistive technology capabilities, they also have the potential to perpetuate, reify and even ...", "subpage_snippet": "", "source": "www.acm.org", "link": "https://www.acm.org/binaries/content/assets/sigs/volunteer_resources/sig-governing-board-annual-reports/fy-19-annual-report/sigs-fy19-appendixd.docx", "content": "While machine learning techniques are enabling breakthroughs in assistive technology capabilities, they also have the potential to perpetuate, reify and even ..."}
+{"idx": 3, "title": "Graph neural networks for predicting metal–ligand ...", "date": "", "ddg_snippet": "by J Toney · 2024 · Cited by 3 — Using this dataset , we train separate graph neural network models to predict the total number and individual identities of ligand coordinating atoms with high ... 43 pages", "subpage_snippet": "", "source": "chemrxiv.org", "link": "https://chemrxiv.org/engage/api-gateway/chemrxiv/assets/orp/resource/item/66fdb15251558a15efe0557a/original/graph-neural-networks-for-predicting-metal-ligand-coordination-of-transition-metal-complexes.pdf", "content": "by J Toney · 2024 · Cited by 3 — Using this dataset , we train separate graph neural network models to predict the total number and individual identities of ligand coordinating atoms with high ... 43 pages"}
+{"idx": 4, "title": "School of Computer Science Courses < Carnegie Mellon ...", "date": "", "ddg_snippet": "Examples are drawn from algorithms, complexity theory, game theory, probability theory, graph theory, automata theory, algebra , cryptography, and combinatorics .", "subpage_snippet": "", "source": "coursecatalog.web.cmu.edu", "link": "http://coursecatalog.web.cmu.edu/schools-colleges/schoolofcomputerscience/courses/", "content": "Examples are drawn from algorithms, complexity theory, game theory, probability theory, graph theory, automata theory, algebra , cryptography, and combinatorics ."}
+{"idx": 5, "title": "", "date": "", "ddg_snippet": "", "subpage_snippet": "", "source": "", "link": "", "content": ""}
+{"idx": 6, "title": "Probabilistic and Statistical Modeling in Computer Science", "date": "", "ddg_snippet": "Dr. Norm Matloff is a professor of computer science at the University of California at Davis, and was formerly a professor of statistics at that university. 553 pages", "subpage_snippet": "", "source": "epsy630.bryer.org", "link": "https://epsy630.bryer.org/materials/Textbooks/ProbStatBook.pdf", "content": "Dr. Norm Matloff is a professor of computer science at the University of California at Davis, and was formerly a professor of statistics at that university. 553 pages"}
+{"idx": 7, "title": "Artificial Intelligence", "date": "", "ddg_snippet": "This Instructor's Solution Manual provides solutions (or at least solution sketches) for almost all of the 400 exercises in Artificial Intelligence : A Modern ...", "subpage_snippet": "", "source": "dl.icdst.org", "link": "https://dl.icdst.org/pdfs/files4/de6f841e0f7d48111fa47bad29356151.pdf", "content": "This Instructor's Solution Manual provides solutions (or at least solution sketches) for almost all of the 400 exercises in Artificial Intelligence : A Modern ..."}
+{"idx": 8, "title": "3-540-45884-0.pdf", "date": "", "ddg_snippet": "This volume contains the research reports of the Discovery Science project in. Japan (No. 10143106), in which more than 60 scientists participated. It was a.", "subpage_snippet": "", "source": "link.springer.com", "link": "https://link.springer.com/content/pdf/10.1007/3-540-45884-0.pdf", "content": "This volume contains the research reports of the Discovery Science project in. Japan (No. 10143106), in which more than 60 scientists participated. It was a."}
+{"idx": 9, "title": "Official Journal of the Bernoulli Society for ...", "date": "", "ddg_snippet": "2 May 2025 — CONTENTS. 843. SPOKOINY, V. and PANOV, M. Accuracy of Gaussian approximation for high-dimensional posterior distributions. 101 pages", "subpage_snippet": "", "source": "www.imstat.org", "link": "https://www.imstat.org/publications/bej/bej_31_2/bej_31_2.pdf", "content": "2 May 2025 — CONTENTS. 843. SPOKOINY, V. and PANOV, M. Accuracy of Gaussian approximation for high-dimensional posterior distributions. 101 pages"}
diff --git a/data/sampled_jsons/SDXL_four_activations_are_selected_diffusion_model_attention_query_key_year_2024.jsonl b/data/sampled_jsons/SDXL_four_activations_are_selected_diffusion_model_attention_query_key_year_2024.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..4c85e399b9cbc0bf517a7493ccce5c340bc34bdb
--- /dev/null
+++ b/data/sampled_jsons/SDXL_four_activations_are_selected_diffusion_model_attention_query_key_year_2024.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "ToMA: Token Merge with Attention for Diffusion Models", "date": "", "ddg_snippet": "Diffusion models excel in high-fidelity image generation but face scalability limits due to transformers’ quadratic attention complexity.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2509.10918v1", "content": "Diffusion models excel in high-fidelity image generation but face scalability limits due to transformers’ quadratic attention complexity."}
+{"idx": 1, "title": "GenTron: Diffusion Transformers for Image and Video Generation", "date": "", "ddg_snippet": "Specifically, our starting point is the foundational work known as DiT [ 45 ] , which introduced a class -conditioned latent diffusion model that ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2312.04557v2", "content": "Specifically, our starting point is the foundational work known as DiT [ 45 ] , which introduced a class -conditioned latent diffusion model that ..."}
+{"idx": 2, "title": "Towards Multimodal Understanding via Stable Diffusion as a", "date": "", "ddg_snippet": "... key limitation is the use of CLIP as the visual encoder; while it can capture coarse global information, it often can miss fine- grained details that ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2507.07106v1", "content": "... key limitation is the use of CLIP as the visual encoder; while it can capture coarse global information, it often can miss fine- grained details that ..."}
+{"idx": 3, "title": "A Survey on Diffusion Language Models", "date": "", "ddg_snippet": "... AGI) have been largely driven by the emergence of autoregressive large language models (LLMs) [ 1 , 2 , 3 , 4 , 5 , 6 , 7 ] and diffusion models ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2508.10875v1", "content": "... AGI) have been largely driven by the emergence of autoregressive large language models (LLMs) [ 1 , 2 , 3 , 4 , 5 , 6 , 7 ] and diffusion models ..."}
+{"idx": 4, "title": "GitHub - apple/ml-stable-diffusion: Stable Diffusion with Core", "date": "", "ddg_snippet": "Performance may vary across different versions of Stable Diffusion due to architecture changes in the model itself.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/apple/ml-stable-diffusion", "content": "Performance may vary across different versions of Stable Diffusion due to architecture changes in the model itself."}
+{"idx": 5, "title": "Aman's AI Journal • Primers • Diffusion Models", "date": "", "ddg_snippet": "Diffusion probabilistic models (also simply called diffusion models ) are generative models , meaning that they are used to generate data similar to ...", "subpage_snippet": "", "source": "aman.ai", "link": "https://aman.ai/primers/ai/diffusion-models/", "content": "Diffusion probabilistic models (also simply called diffusion models ) are generative models , meaning that they are used to generate data similar to ..."}
+{"idx": 6, "title": "Detail++: Training-Free Detail Enhancer for Text-to-Image", "date": "", "ddg_snippet": "Diffusion Models have become the dominant approach in text-to-image (T2I) generation [ 2 , 42 , 45 , 48 , 12 , 27 , 13 ] .", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2507.17853v1", "content": "Diffusion Models have become the dominant approach in text-to-image (T2I) generation [ 2 , 42 , 45 , 48 , 12 , 27 , 13 ] ."}
+{"idx": 7, "title": "Installation on Apple Silicon ·", "date": "", "ddg_snippet": "Most samplers are known to work with the only exception being the PLMS sampler when using the Stable Diffusion 2.0 model .", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Installation-on-Apple-Silicon", "content": "Most samplers are known to work with the only exception being the PLMS sampler when using the Stable Diffusion 2.0 model ."}
+{"idx": 8, "title": "Metrics — OpenVINO™ documentation", "date": "", "ddg_snippet": "PixArt-α: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis with OpenVINO ... Models with OpenVINO™ and 4th Gen ...", "subpage_snippet": "", "source": "docs.openvino.ai", "link": "https://docs.openvino.ai/2024/openvino-workflow/model-server/ovms_docs_metrics.html", "content": "PixArt-α: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis with OpenVINO ... Models with OpenVINO™ and 4th Gen ..."}
+{"idx": 9, "title": "Heterogeneous Execution — OpenVINO™ documentation", "date": "", "ddg_snippet": "PixArt-α: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis with OpenVINO ... Models with OpenVINO™ and 4th Gen ...", "subpage_snippet": "", "source": "docs.openvino.ai", "link": "https://docs.openvino.ai/2024/openvino-workflow/running-inference/inference-devices-and-modes/hetero-execution.html", "content": "PixArt-α: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis with OpenVINO ... Models with OpenVINO™ and 4th Gen ..."}
diff --git a/data/sampled_jsons/Statistical_Collusion_by_Collectives_on_Learning_Platforms_Hoeffding_error_term_Rs(k).jsonl b/data/sampled_jsons/Statistical_Collusion_by_Collectives_on_Learning_Platforms_Hoeffding_error_term_Rs(k).jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..a90f9f4b0204cbc0f6d734ef6512f53d7ee7402e
--- /dev/null
+++ b/data/sampled_jsons/Statistical_Collusion_by_Collectives_on_Learning_Platforms_Hoeffding_error_term_Rs(k).jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Statistical Collusion by Collectives on Learning Platforms", "date": "", "ddg_snippet": "7 Feb 2025 — Throughout this paper, we will use Hoeffding's concentration inequality (Lemma D.1) for simplicity. We will denote Hoeffding error terms as ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2502.04879v1", "content": "7 Feb 2025 — Throughout this paper, we will use Hoeffding's concentration inequality (Lemma D.1) for simplicity. We will denote Hoeffding error terms as ..."}
+{"idx": 1, "title": "Improving Human Integration across the Machine Learning ...", "date": "", "ddg_snippet": "by C Rastogi · 2024 · Cited by 2 — Siva's love for and expertise in statistical learning theory reflects in his teaching-oriented approach to advising, and his clarity in conveying fundamental. 263 pages", "subpage_snippet": "", "source": "ml.cmu.edu", "link": "https://ml.cmu.edu/research/phd-dissertation-pdfs/thesis_rastogi_charvi.pdf", "content": "by C Rastogi · 2024 · Cited by 2 — Siva's love for and expertise in statistical learning theory reflects in his teaching-oriented approach to advising, and his clarity in conveying fundamental. 263 pages"}
+{"idx": 2, "title": "Machine Learning and Knowledge Discovery in Databases", "date": "", "ddg_snippet": "9 Sept 2024 — The annual ECML PKDD conference acts as a world-wide platform showcasing the latest advancements in machine learning and knowledge discovery in ...", "subpage_snippet": "", "source": "link.springer.com", "link": "https://link.springer.com/content/pdf/10.1007/978-3-031-70341-6.pdf", "content": "9 Sept 2024 — The annual ECML PKDD conference acts as a world-wide platform showcasing the latest advancements in machine learning and knowledge discovery in ..."}
+{"idx": 3, "title": "Computer Science", "date": "", "ddg_snippet": "6 days ago — Statistical model checking (SMC) randomly samples probabilistic models to approximate quantities of interest with statistical error guarantees.", "subpage_snippet": "", "source": "www.arxiv.org", "link": "https://www.arxiv.org/list/cs/new?skip=550&show=1000", "content": "6 days ago — Statistical model checking (SMC) randomly samples probabilistic models to approximate quantities of interest with statistical error guarantees."}
+{"idx": 4, "title": "UC Berkeley", "date": "", "ddg_snippet": "by EV Mazumdar · 2021 — Supposing Algorithm 1 achieves a sufficiently small training error ε > 0, the second term above can be bounded above by a calculable small amount which we ...", "subpage_snippet": "", "source": "escholarship.org", "link": "https://escholarship.org/content/qt8h17j4jb/qt8h17j4jb.pdf", "content": "by EV Mazumdar · 2021 — Supposing Algorithm 1 achieves a sufficiently small training error ε > 0, the second term above can be bounded above by a calculable small amount which we ..."}
+{"idx": 5, "title": "IJCAI'17 Program Schedule", "date": "", "ddg_snippet": "... statistical learning methods may be affected by the problems of word sparseness and synonyms. Although recent progress in neural word embedding methods have ...", "subpage_snippet": "", "source": "static.ijcai.org", "link": "https://static.ijcai.org/2017-Program.html", "content": "... statistical learning methods may be affected by the problems of word sparseness and synonyms. Although recent progress in neural word embedding methods have ..."}
+{"idx": 6, "title": "Advanced Mathematical Methods in Intelligent Multimedia", "date": "", "ddg_snippet": "... error triplet expansion; prediction- error histogram. 1. Introduction. With the rapid development of computer and communication technologies, the pri- vacy ...", "subpage_snippet": "", "source": "mdpi-res.com", "link": "https://mdpi-res.com/bookfiles/book/7935/Advanced_Mathematical_Methods_in_Intelligent_Multimedia_Security_and_Applications.pdf?v=1743127733", "content": "... error triplet expansion; prediction- error histogram. 1. Introduction. With the rapid development of computer and communication technologies, the pri- vacy ..."}
+{"idx": 7, "title": "Spectral Methods for Social Media Data Analysis by Fan ...", "date": "", "ddg_snippet": "by F Chen · 2021 · Cited by 2 — Second, in statistical models with k true dimensions, bias differentially effects the k and k +1 sample eigenvalues, and this bias blurs any gap or elbow ... 287 pages", "subpage_snippet": "", "source": "asset.library.wisc.edu", "link": "https://asset.library.wisc.edu/1711.dl/YFGOBWPGELYL69D/R/file-8d5d2.pdf", "content": "by F Chen · 2021 · Cited by 2 — Second, in statistical models with k true dimensions, bias differentially effects the k and k +1 sample eigenvalues, and this bias blurs any gap or elbow ... 287 pages"}
+{"idx": 8, "title": "Algorithms for Games AI", "date": "", "ddg_snippet": "Games have long been benchmarks for AI algorithms and, with the boost of computa- tional power and the application of new algorithms, AI systems have ...", "subpage_snippet": "", "source": "mdpi-res.com", "link": "https://mdpi-res.com/bookfiles/book/11350/Algorithms_for_Games_AI.pdf?v=1755579679", "content": "Games have long been benchmarks for AI algorithms and, with the boost of computa- tional power and the application of new algorithms, AI systems have ..."}
+{"idx": 9, "title": "Computer Science & Information Technology 210 Machine ...", "date": "", "ddg_snippet": "Committees rigorously invited submissions for many months from researchers, scientists, engineers, students and practitioners related to the relevant themes and ...", "subpage_snippet": "", "source": "aircconline.com", "link": "https://aircconline.com/csit/csit1302.pdf", "content": "Committees rigorously invited submissions for many months from researchers, scientists, engineers, students and practitioners related to the relevant themes and ..."}
diff --git a/data/sampled_jsons/TCE_Transformer_Context_Encoder_Li_et_al_2024_reinforcement_learning.jsonl b/data/sampled_jsons/TCE_Transformer_Context_Encoder_Li_et_al_2024_reinforcement_learning.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..dd903dd5b98b7aad6dfe104580178e781b1b7947
--- /dev/null
+++ b/data/sampled_jsons/TCE_Transformer_Context_Encoder_Li_et_al_2024_reinforcement_learning.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "On Transforming Reinforcement Learning With Transformers: The ...", "date": "", "ddg_snippet": "Transformers , originally devised for natural language processing (NLP), have also produced significant successes in computer vision (CV). Due to their strong expression power, researchers are investigating ways to deploy transformers for reinforcement learning (RL), and transformer -based models have manifested their potential in representative RL benchmarks. In this paper, we collect and ...", "subpage_snippet": "", "source": "ieeexplore.ieee.org", "link": "https://ieeexplore.ieee.org/abstract/document/10546317", "content": "Transformers , originally devised for natural language processing (NLP), have also produced significant successes in computer vision (CV). Due to their strong expression power, researchers are investigating ways to deploy transformers for reinforcement learning (RL), and transformer -based models have manifested their potential in representative RL benchmarks. In this paper, we collect and ..."}
+{"idx": 1, "title": "Application of Transformer for Encoding States in Reinforcement Learning", "date": "", "ddg_snippet": "This is also supported by studies [9], which demonstrate the effectiveness of transformer encoders as policies in variable action environments. In addition, the author of [10] proposed a new perspective on transformers as meta- learning tools in the context of deep reinforcement learning , emphasizing their versatility and adaptability.", "subpage_snippet": "", "source": "link.springer.com", "link": "https://link.springer.com/article/10.3103/S8756699024700705", "content": "This is also supported by studies [9], which demonstrate the effectiveness of transformer encoders as policies in variable action environments. In addition, the author of [10] proposed a new perspective on transformers as meta- learning tools in the context of deep reinforcement learning , emphasizing their versatility and adaptability."}
+{"idx": 2, "title": "[2307.05979] Transformers in Reinforcement Learning: A Survey", "date": "", "ddg_snippet": "Transformers have significantly impacted domains like natural language processing, computer vision, and robotics, where they improve performance compared to other neural networks. This survey explores how transformers are used in reinforcement learning (RL), where they are seen as a promising solution for addressing challenges such as unstable training, credit assignment, lack of ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2307.05979", "content": "Transformers have significantly impacted domains like natural language processing, computer vision, and robotics, where they improve performance compared to other neural networks. This survey explores how transformers are used in reinforcement learning (RL), where they are seen as a promising solution for addressing challenges such as unstable training, credit assignment, lack of ..."}
+{"idx": 3, "title": "Contextual Transformers for Goal-Oriented Reinforcement Learning ...", "date": "", "ddg_snippet": "Transformer architectures have become popular across deep- learning disciplines due to their capability of efficiently integrating information across extensive temporal spans and handling large datasets. Recently, this property of transformer models has also been utilized for reinforcement learning (RL) by learning in- context . In in- context learning for decision-making problems, i.e., RL, a ...", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/10.1007/978-3-031-77915-2_15", "content": "Transformer architectures have become popular across deep- learning disciplines due to their capability of efficiently integrating information across extensive temporal spans and handling large datasets. Recently, this property of transformer models has also been utilized for reinforcement learning (RL) by learning in- context . In in- context learning for decision-making problems, i.e., RL, a ..."}
+{"idx": 4, "title": "A Survey on Transformers in Reinforcement Learning", "date": "", "ddg_snippet": "Transformer has been considered the dominating neural architecture in NLP and CV, mostly under supervised settings. Recently, a similar surge of using Transformers has appeared in the domain of reinforcement learning (RL), but it is faced with unique design choices and challenges brought by the nature of RL. However, the evolution of Transformers in RL has not yet been well unraveled. In this ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2301.03044", "content": "Transformer has been considered the dominating neural architecture in NLP and CV, mostly under supervised settings. Recently, a similar surge of using Transformers has appeared in the domain of reinforcement learning (RL), but it is faced with unique design choices and challenges brought by the nature of RL. However, the evolution of Transformers in RL has not yet been well unraveled. In this ..."}
+{"idx": 5, "title": "Improving the performance of Transformer Context Encoders for NER", "date": "", "ddg_snippet": "In this paper, we compare the performance of the Transformer and Recurrent architecture as context encoders on the Named Entity Recognition (NER) task. We vary the character-level representation module from the previously proposed NER models in literature and show how the modification can improve the NER model's performance.", "subpage_snippet": "", "source": "ieeexplore.ieee.org", "link": "https://ieeexplore.ieee.org/document/9627061", "content": "In this paper, we compare the performance of the Transformer and Recurrent architecture as context encoders on the Named Entity Recognition (NER) task. We vary the character-level representation module from the previously proposed NER models in literature and show how the modification can improve the NER model's performance."}
+{"idx": 6, "title": "On Transforming Reinforcement Learning With Transformers: The ...", "date": "", "ddg_snippet": "Due to their strong expression power, researchers are investigating ways to deploy transformers for reinforcement learning (RL), and transformer -based models have manifested their potential in representative RL benchmarks.", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/10.1109/TPAMI.2024.3408271", "content": "Due to their strong expression power, researchers are investigating ways to deploy transformers for reinforcement learning (RL), and transformer -based models have manifested their potential in representative RL benchmarks."}
+{"idx": 7, "title": "On Transforming Reinforcement Learning by Transformer: The Development ...", "date": "", "ddg_snippet": "Transformer , originally devised for natural language processing, has also attested significant success in computer vision. Thanks to its super expressive power, researchers are investigating ways to deploy transformers to reinforcement learning (RL) and the transformer -based models have manifested their potential in representative RL benchmarks. In this paper, we collect and dissect recent ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2212.14164", "content": "Transformer , originally devised for natural language processing, has also attested significant success in computer vision. Thanks to its super expressive power, researchers are investigating ways to deploy transformers to reinforcement learning (RL) and the transformer -based models have manifested their potential in representative RL benchmarks. In this paper, we collect and dissect recent ..."}
+{"idx": 8, "title": "Enhancing Reinforcement Learning via Transformer-Based State Predictive ...", "date": "", "ddg_snippet": "Enhancing state representations can effectively mitigate the issue of low sample efficiency in reinforcement learning (RL) within high-dimensional input environments. Existing methods attempt to improve sample efficiency by learning predictive state representations from sequence data. However, there still remain significant challenges in achieving a comprehensive understanding and learning of ...", "subpage_snippet": "", "source": "ieeexplore.ieee.org", "link": "https://ieeexplore.ieee.org/abstract/document/10477774", "content": "Enhancing state representations can effectively mitigate the issue of low sample efficiency in reinforcement learning (RL) within high-dimensional input environments. Existing methods attempt to improve sample efficiency by learning predictive state representations from sequence data. However, there still remain significant challenges in achieving a comprehensive understanding and learning of ..."}
+{"idx": 9, "title": "On Transforming Reinforcement Learning with Transformers: The ...", "date": "", "ddg_snippet": "1 INTRODUCTION R ECENTLY, the transformer architecture has made sub- stantial progress in natural language processing (NLP) tasks [1]. For example, generative pretraining (GPT) series models [2] and bidirectional encoder representations from transformers (BERT) models [3] have achieved state-of-the-art performance on a wide range of downstream tasks (e.g. question answering (QA) and sentence ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2212.14164", "content": "1 INTRODUCTION R ECENTLY, the transformer architecture has made sub- stantial progress in natural language processing (NLP) tasks [1]. For example, generative pretraining (GPT) series models [2] and bidirectional encoder representations from transformers (BERT) models [3] have achieved state-of-the-art performance on a wide range of downstream tasks (e.g. question answering (QA) and sentence ..."}
diff --git a/data/sampled_jsons/Taming_Knowledge_Conflicts_in_Language_Models_GitHub_JUICE_World_Capital_Table_3.jsonl b/data/sampled_jsons/Taming_Knowledge_Conflicts_in_Language_Models_GitHub_JUICE_World_Capital_Table_3.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..090da3ee61f76fd167132e178104ad95ed47d2ce
--- /dev/null
+++ b/data/sampled_jsons/Taming_Knowledge_Conflicts_in_Language_Models_GitHub_JUICE_World_Capital_Table_3.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Taming Knowledge Conflicts in Language Models - GitHub", "date": "", "ddg_snippet": "This repository contains the code and data of the ICML 25 Spotlight Paper Taming Knowledge Conflicts in Language Models . The code is now still being updated.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/GaotangLi/JUICE", "content": "This repository contains the code and data of the ICML 25 Spotlight Paper Taming Knowledge Conflicts in Language Models . The code is now still being updated."}
+{"idx": 1, "title": "Taming Knowledge Conflict in Language Models - gaotangli.github.io", "date": "", "ddg_snippet": "Language Models (LMs) often encounter knowledge conflicts when parametric memory contradicts contextual knowledge . Previous works attribute this conflict to the interplay between \"memory heads\" and \"context heads\", attention heads assumed to promote either memory or context exclusively. In this study, we go beyond this fundamental assumption by uncovering a critical phenomenon we term the ...", "subpage_snippet": "", "source": "gaotangli.github.io", "link": "https://gaotangli.github.io/project_page/Taming-Knowledge-Conflict/", "content": "Language Models (LMs) often encounter knowledge conflicts when parametric memory contradicts contextual knowledge . Previous works attribute this conflict to the interplay between \"memory heads\" and \"context heads\", attention heads assumed to promote either memory or context exclusively. In this study, we go beyond this fundamental assumption by uncovering a critical phenomenon we term the ..."}
+{"idx": 2, "title": "Taming Knowledge Conflicts in Language Models - arXiv.org", "date": "", "ddg_snippet": "Abstract Language Models (LMs) often encounter knowl-edge conflicts when parametric memory con-tradicts contextual knowledge . Previous works attribute this conflict to the interplay between \"memory heads\" and \"context heads\", attention heads assumed to promote either memory or con-text exclusively. In this study, we go beyond this fundamental assumption by uncovering a criti-cal ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2503.10996v1", "content": "Abstract Language Models (LMs) often encounter knowl-edge conflicts when parametric memory con-tradicts contextual knowledge . Previous works attribute this conflict to the interplay between \"memory heads\" and \"context heads\", attention heads assumed to promote either memory or con-text exclusively. In this study, we go beyond this fundamental assumption by uncovering a criti-cal ..."}
+{"idx": 3, "title": "ICML Poster Taming Knowledge Conflicts in Language Models", "date": "", "ddg_snippet": "Language models frequently encounter \" knowledge conflicts ,\" where their pre-trained knowledge contradicts the information provided by specific contexts. These conflicts often arise in context-dependent systems, such as retrieval-augmented generation and tools integrated with language models .", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/poster/46677", "content": "Language models frequently encounter \" knowledge conflicts ,\" where their pre-trained knowledge contradicts the information provided by specific contexts. These conflicts often arise in context-dependent systems, such as retrieval-augmented generation and tools integrated with language models ."}
+{"idx": 4, "title": "Taming Knowledge Conflicts in Language Models - Semantic Scholar", "date": "", "ddg_snippet": "This work proposes Just Run Twice ( JuICE ), a test-time attention intervention method that steers LMs toward either parametric beliefs or contextual knowledge without requiring fine-tuning, and identifies a set of reliable attention heads and leverages a dual-run approach to mitigate the superposition effects. Language Models (LMs) often encounter knowledge conflicts when parametric memory ...", "subpage_snippet": "", "source": "www.semanticscholar.org", "link": "https://www.semanticscholar.org/paper/Taming-Knowledge-Conflicts-in-Language-Models-Li-Chen/b7ba9df4eb239708cf48f25be87b5bceeca010e3", "content": "This work proposes Just Run Twice ( JuICE ), a test-time attention intervention method that steers LMs toward either parametric beliefs or contextual knowledge without requiring fine-tuning, and identifies a set of reliable attention heads and leverages a dual-run approach to mitigate the superposition effects. Language Models (LMs) often encounter knowledge conflicts when parametric memory ..."}
+{"idx": 5, "title": "Knowledge Conflict - a gaotang Collection - Hugging Face", "date": "", "ddg_snippet": "Parametric dataset related to the paper \" Taming Knowledge Conflict in Language Models \".", "subpage_snippet": "", "source": "huggingface.co", "link": "https://huggingface.co/collections/gaotang/knowledge-conflict-681408090438726d04c440ca", "content": "Parametric dataset related to the paper \" Taming Knowledge Conflict in Language Models \"."}
+{"idx": 6, "title": "[2503.10996] Taming Knowledge Conflicts in Language Models", "date": "", "ddg_snippet": "Language Models (LMs) often encounter knowledge conflicts when parametric memory contradicts contextual knowledge . Previous works attribute this conflict to the interplay between \"memory heads\" and \"context heads\", attention heads assumed to promote either memory or context exclusively. In this study, we go beyond this fundamental assumption by uncovering a critical phenomenon we term the ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2503.10996", "content": "Language Models (LMs) often encounter knowledge conflicts when parametric memory contradicts contextual knowledge . Previous works attribute this conflict to the interplay between \"memory heads\" and \"context heads\", attention heads assumed to promote either memory or context exclusively. In this study, we go beyond this fundamental assumption by uncovering a critical phenomenon we term the ..."}
+{"idx": 7, "title": "Resolving Knowledge Conflicts in Large Language Models - GitHub", "date": "", "ddg_snippet": "The top-level keys in the json file correspond to primary fields, and each data point within a field is represented as a dictionary, with the following key-value pairs: main_entity (str): an entity from the generated entity list parametric_knowledge (str): extracted parametric knowledge about the main_entity named_entity_lst (lst): named entities with corresponding types returned by NER models ...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/yikee/Knowledge_Conflict", "content": "The top-level keys in the json file correspond to primary fields, and each data point within a field is represented as a dictionary, with the following key-value pairs: main_entity (str): an entity from the generated entity list parametric_knowledge (str): extracted parametric knowledge about the main_entity named_entity_lst (lst): named entities with corresponding types returned by NER models ..."}
+{"idx": 8, "title": "ICML Taming Knowledge Conflicts in Language Models", "date": "", "ddg_snippet": "Poster in Workshop: Actionable Interpretability Taming Knowledge Conflicts in Language Models Gaotang Li · Yuzhong Chen · Hanghang Tong [ Abstract ] [ Project Page ] [ OpenReview] Sat 19 Jul 10:40 a.m. PDT — 11:40 a.m. PDT", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/49596", "content": "Poster in Workshop: Actionable Interpretability Taming Knowledge Conflicts in Language Models Gaotang Li · Yuzhong Chen · Hanghang Tong [ Abstract ] [ Project Page ] [ OpenReview] Sat 19 Jul 10:40 a.m. PDT — 11:40 a.m. PDT"}
+{"idx": 9, "title": "Taming Knowledge Conflicts in Language Models | alphaXiv", "date": "", "ddg_snippet": "View recent discussion. Abstract: Language Models (LMs) often encounter knowledge conflicts when parametric memory contradicts contextual knowledge . Previous works attribute this conflict to the interplay between \"memory heads\" and \"context heads\", attention heads assumed to promote either memory or context exclusively. In this study, we go beyond this fundamental assumption by uncovering a ...", "subpage_snippet": "", "source": "www.alphaxiv.org", "link": "https://www.alphaxiv.org/overview/2503.10996v2", "content": "View recent discussion. Abstract: Language Models (LMs) often encounter knowledge conflicts when parametric memory contradicts contextual knowledge . Previous works attribute this conflict to the interplay between \"memory heads\" and \"context heads\", attention heads assumed to promote either memory or context exclusively. In this study, we go beyond this fundamental assumption by uncovering a ..."}
diff --git a/data/sampled_jsons/Training-Free_Diffusion_Model_Alignment_with_Sampling_Demons_paper_Tanh_vs_Tanh-C.jsonl b/data/sampled_jsons/Training-Free_Diffusion_Model_Alignment_with_Sampling_Demons_paper_Tanh_vs_Tanh-C.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..6e23b3d83b53a7a82d7216dad750ed3c24cbeeca
--- /dev/null
+++ b/data/sampled_jsons/Training-Free_Diffusion_Model_Alignment_with_Sampling_Demons_paper_Tanh_vs_Tanh-C.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Training-free Diffusion Model Alignment with Sampling Demons DemonSampling/README.md at main · aiiu-lab ... - GitHub Published as a conference paper at ICLR 2025 - OpenReview Training-Free Diffusion Model Alignment with Sampling Demons Training-free Diffusion Model Alignment with Sampling Demons ... Training-free Diffusion Model Alignment with Sampling Demons Training - free Diffusion Model Alignment with Sampling Demons Training-free Diffusion Model Alignment with Sampling Demons - arXi… Training-free Diffusion Model Alignment with Sampling Demons - arXi… Training-free Diffusion Model Alignment with Sampling Demons - arXi… Training-free Diffusion Model Alignment with Sampling Demons - arXi… Training-free Diffusion Model Alignment with Sampling Demons - arXi… Training-free Diffusion Model Alignment with Sampling Demons", "date": "", "ddg_snippet": "Oct 8, 2024 · To the best of our knowledge, the proposed approach is the first inference-time, backpropagation- free preference alignment method for diffusion models . Our method can be easily integrated with existing diffusion models without further training. This repository contains the official implementation of Sampling Demon , an inference-time, backpropagation- free preference alignment method for diffusion models . To the best of our knowledge, the proposed approach is the first inference-time, backpropagation- free preference alignment method for diffusion models . Our method can be easily integrated with existing diffusion models without further training . Aligning diffusion models with user preferences has been a key challenge.Existing methods for aligning diffusion models either require retraining or are limited to differentiable reward functions.To address these limitations, we propose a stochastic optimization approach, dubbed Demon, to guide the denoising process at inference time without ... This paper introduces a training - free technique called \" Sampling Demons \" for aligning diffusion models with target objectives. The method works by modifying the diffusion sampling process to match the desired output during inference, without requiring any additional training. This work proposes a stochastic optimization approach, dubbed Demon, to guide the denoising process at inference time without backpropagation through reward functions or model retraining, which is the first inference-time, backpropagation-free preference alignment method for diffusion models . Can diffusion models be aligned with user preferences? Aligning diffusion models with user preferences has been a key challenge . Existing methods for aligning diffusion models either require retraining or are limited to differentiable reward functions. Can diffusion models be trained with reinforcement learning? Training diffusion models with reinforcement learning. In International Conference on Machine Learning, 2023. Clark et al. (2024) Kevin Clark, Paul Vicol, Kevin Swersky, and David J Fleet. Directly fine-tuning diffusion models on differentiable rewards. Do diffusion models beat Gans on image synthesis? Diffusion models beat GANs on image synthesis . In Advances in Neural Information Processing Systems, 2021. Fan et al. (2023) Ying Fan, Olivia Watkins, Yuqing Du, Hao Liu, Moonkyung Ryu, Craig Boutilier, Pieter Abbeel, Mohammad Ghavamzadeh, Kangwook Lee, and Kimin Lee. What is a diffusion model? Diffusion models have been the state-of-the-art for image generation (Sohl-Dickstein et al., 2015; Ho et al., 2020; Song et al., 2021; Karras et al., 2022; Saharia et al., 2022; Rombach et al., 2022), but, commonly, the end users’ preferences and intention diverge from the data distribution on which the model was trained. Does stable diffusion improve aesthetics score? We demonstrate that our approach significantly improves the average aesthetics score (LAION, 2023) of Stable Diffusion models, achieving averages well above 8.0 compared to the Best-of-N random sampling upper bounds of 6.5 for SD v1.4 and 7 for SDXL. Can human feedback be used to fine-tune diffusion models without reward models? Using human feedback to fine-tune diffusion models without any reward model , 2024. Although we keep the main paper self-consistent, we provide this section to establish a consistent notation and convention for this paper as an aid. Instead of just ODE, we use PF-ODE to highlight Song et al. (2021) ’s contribution or when the context is unclear. To the best of our knowledge, the proposed approach is the first inference-time, backpropagation- free preference alignment method for diffusion models. Our method can be easily integrated with existing diffusion models without further training.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2410.05760", "content": "Oct 8, 2024 · To the best of our knowledge, the proposed approach is the first inference-time, backpropagation- free preference alignment method for diffusion models . Our method can be easily integrated with existing diffusion models without further training. This repository contains the official implementation of Sampling Demon , an inference-time, backpropagation- free preference alignment method for diffusion models . To the best of our knowledge, the proposed approach is the first inference-time, backpropagation- free preference alignment method for diffusion models . Our method can be easily integrated with existing diffusion models without further training . Aligning diffusion models with user preferences has been a key challenge.Existing methods for aligning diffusion models either require retraining or are limited to differentiable reward functions.To address these limitations, we propose a stochastic optimization approach, dubbed Demon, to guide the denoising process at inference time without ... This paper introduces a training - free technique called \" Sampling Demons \" for aligning diffusion models with target objectives. The method works by modifying the diffusion sampling process to match the desired output during inference, without requiring any additional training. This work proposes a stochastic optimization approach, dubbed Demon, to guide the denoising process at inference time without backpropagation through reward functions or model retraining, which is the first inference-time, backpropagation-free preference alignment method for diffusion models . Can diffusion models be aligned with user preferences? Aligning diffusion models with user preferences has been a key challenge . Existing methods for aligning diffusion models either require retraining or are limited to differentiable reward functions. Can diffusion models be trained with reinforcement learning? Training diffusion models with reinforcement learning. In International Conference on Machine Learning, 2023. Clark et al. (2024) Kevin Clark, Paul Vicol, Kevin Swersky, and David J Fleet. Directly fine-tuning diffusion models on differentiable rewards. Do diffusion models beat Gans on image synthesis? Diffusion models beat GANs on image synthesis . In Advances in Neural Information Processing Systems, 2021. Fan et al. (2023) Ying Fan, Olivia Watkins, Yuqing Du, Hao Liu, Moonkyung Ryu, Craig Boutilier, Pieter Abbeel, Mohammad Ghavamzadeh, Kangwook Lee, and Kimin Lee. What is a diffusion model? Diffusion models have been the state-of-the-art for image generation (Sohl-Dickstein et al., 2015; Ho et al., 2020; Song et al., 2021; Karras et al., 2022; Saharia et al., 2022; Rombach et al., 2022), but, commonly, the end users’ preferences and intention diverge from the data distribution on which the model was trained. Does stable diffusion improve aesthetics score? We demonstrate that our approach significantly improves the average aesthetics score (LAION, 2023) of Stable Diffusion models, achieving averages well above 8.0 compared to the Best-of-N random sampling upper bounds of 6.5 for SD v1.4 and 7 for SDXL. Can human feedback be used to fine-tune diffusion models without reward models? Using human feedback to fine-tune diffusion models without any reward model , 2024. Although we keep the main paper self-consistent, we provide this section to establish a consistent notation and convention for this paper as an aid. Instead of just ODE, we use PF-ODE to highlight Song et al. (2021) ’s contribution or when the context is unclear. To the best of our knowledge, the proposed approach is the first inference-time, backpropagation- free preference alignment method for diffusion models. Our method can be easily integrated with existing diffusion models without further training."}
+{"idx": 1, "title": "DemonSampling/README.md at main · aiiu-lab ... - GitHub", "date": "", "ddg_snippet": "This repository contains the official implementation of Sampling Demon , an inference-time, backpropagation- free preference alignment method for diffusion models .", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/aiiu-lab/DemonSampling/blob/main/README.md", "content": "This repository contains the official implementation of Sampling Demon , an inference-time, backpropagation- free preference alignment method for diffusion models ."}
+{"idx": 2, "title": "Published as a conference paper at ICLR 2025 - OpenReview", "date": "", "ddg_snippet": "To the best of our knowledge, the proposed approach is the first inference-time, backpropagation- free preference alignment method for diffusion models . Our method can be easily integrated with existing diffusion models without further training .", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=tfemquulED", "content": "To the best of our knowledge, the proposed approach is the first inference-time, backpropagation- free preference alignment method for diffusion models . Our method can be easily integrated with existing diffusion models without further training ."}
+{"idx": 3, "title": "Training-Free Diffusion Model Alignment with Sampling Demons", "date": "", "ddg_snippet": "Aligning diffusion models with user preferences has been a key challenge.Existing methods for aligning diffusion models either require retraining or are limited to differentiable reward functions.To address these limitations, we propose a stochastic optimization approach, dubbed Demon, to guide the denoising process at inference time without ...", "subpage_snippet": "", "source": "proceedings.iclr.cc", "link": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/eeab2e00835c71d64458ad1821e05664-Abstract-Conference.html", "content": "Aligning diffusion models with user preferences has been a key challenge.Existing methods for aligning diffusion models either require retraining or are limited to differentiable reward functions.To address these limitations, we propose a stochastic optimization approach, dubbed Demon, to guide the denoising process at inference time without ..."}
+{"idx": 4, "title": "Training-free Diffusion Model Alignment with Sampling Demons ...", "date": "", "ddg_snippet": "This paper introduces a training - free technique called \" Sampling Demons \" for aligning diffusion models with target objectives. The method works by modifying the diffusion sampling process to match the desired output during inference, without requiring any additional training.", "subpage_snippet": "", "source": "www.aimodels.fyi", "link": "https://www.aimodels.fyi/papers/arxiv/training-free-diffusion-model-alignment-sampling-demons", "content": "This paper introduces a training - free technique called \" Sampling Demons \" for aligning diffusion models with target objectives. The method works by modifying the diffusion sampling process to match the desired output during inference, without requiring any additional training."}
+{"idx": 5, "title": "Training-free Diffusion Model Alignment with Sampling Demons", "date": "", "ddg_snippet": "This work proposes a stochastic optimization approach, dubbed Demon, to guide the denoising process at inference time without backpropagation through reward functions or model retraining, which is the first inference-time, backpropagation-free preference alignment method for diffusion models .", "subpage_snippet": "", "source": "www.semanticscholar.org", "link": "https://www.semanticscholar.org/paper/Training-free-Diffusion-Model-Alignment-with-Demons-Yeh-Lee/25609eb8a8cc9477ba6c654e40b03928f0a5c8af/figure/0", "content": "This work proposes a stochastic optimization approach, dubbed Demon, to guide the denoising process at inference time without backpropagation through reward functions or model retraining, which is the first inference-time, backpropagation-free preference alignment method for diffusion models ."}
+{"idx": 6, "title": "Training-free Diffusion Model Alignment with Sampling Demons", "date": "", "ddg_snippet": "To the best of our knowledge, the proposed approach is the first inference-time, backpropagation- free preference alignment method for diffusion models. Our method can be easily integrated with existing diffusion models without further training.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2410.05760v1", "content": "To the best of our knowledge, the proposed approach is the first inference-time, backpropagation- free preference alignment method for diffusion models. Our method can be easily integrated with existing diffusion models without further training."}
+{"idx": 7, "title": "Training-Free Diffusion Model Alignment with Sampling ...", "date": "", "ddg_snippet": "by PH Yeh · Cited by 5 — Reviewer RS1X highlighted a performance discrepancy between the ODE solver and the CM solver, resulting in a performance gap between Tanh and Tanh - C . To address ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=tfemquulED", "content": "by PH Yeh · Cited by 5 — Reviewer RS1X highlighted a performance discrepancy between the ODE solver and the CM solver, resulting in a performance gap between Tanh and Tanh - C . To address ..."}
+{"idx": 8, "title": "Training-free Diffusion Model Alignment with Sampling ...", "date": "", "ddg_snippet": "In terms of reward queries, Tanh/Tanh-C outperforms other baseline methods in most cases, including our Boltzmann method and Best-of-N. Our methods are even ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2410.05760v2", "content": "In terms of reward queries, Tanh/Tanh-C outperforms other baseline methods in most cases, including our Boltzmann method and Best-of-N. Our methods are even ..."}
+{"idx": 9, "title": "TRAINING-FREE DIFFUSION MODEL ALIGNMENT WITH ...", "date": "", "ddg_snippet": "The only difference between Tanh - C and Tanh Demon lies in how r ◦ c is implemented. Analysis of the data in Table 2 and Figure 4 indicates that Tanh - C's reward ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf/bcc9f999c2180e515ff2fd9638d84eef96f5f502.pdf", "content": "The only difference between Tanh - C and Tanh Demon lies in how r ◦ c is implemented. Analysis of the data in Table 2 and Figure 4 indicates that Tanh - C's reward ..."}
diff --git a/data/sampled_jsons/Unbiased_Recommender_Learning_from_Implicit_Feedback_via_Weakly_Supervised_Learning_equation_6.jsonl b/data/sampled_jsons/Unbiased_Recommender_Learning_from_Implicit_Feedback_via_Weakly_Supervised_Learning_equation_6.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..167cd02006278765e8bbc6928d4f06ecff8ee698
--- /dev/null
+++ b/data/sampled_jsons/Unbiased_Recommender_Learning_from_Implicit_Feedback_via_Weakly_Supervised_Learning_equation_6.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "PDF Unbiased Recommender Learning from Implicit Feedback via Weakly ...", "date": "", "ddg_snippet": "To address this issue, we introduce WeaklyRec, a model-agnostic framework that reframes implicit feedback recommendation as a weakly supervised learning task, eliminating the need for negative samples. However, its unbiasedness hinges on the accurate estimation of the class prior.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/attachment?id=0E5rZOGA13&name=pdf", "content": "To address this issue, we introduce WeaklyRec, a model-agnostic framework that reframes implicit feedback recommendation as a weakly supervised learning task, eliminating the need for negative samples. However, its unbiasedness hinges on the accurate estimation of the class prior."}
+{"idx": 1, "title": "PDF Unbiased Recommender Learning from Biased Graded Implicit Feedback", "date": "", "ddg_snippet": "To better utilize graded implicit feedback , we for-mulate a recommendation using the MNAR graded implicit feed-back from a statistical estimation perspective, which allows us to theoretically characterize the bias in using the graded implicit feed-back .", "subpage_snippet": "", "source": "decisionmaking4ir.github.io", "link": "https://decisionmaking4ir.github.io/WSDM-2022/papers/Suguru.pdf", "content": "To better utilize graded implicit feedback , we for-mulate a recommendation using the MNAR graded implicit feed-back from a statistical estimation perspective, which allows us to theoretically characterize the bias in using the graded implicit feed-back ."}
+{"idx": 2, "title": "Unbiased Recommender Learning from Biased Graded Implicit Feedback", "date": "", "ddg_snippet": "Binary user-behavior logs such as clicks or views, called implicit feedback , are often used to build recommender systems because of its general availability in real practice. Most existing studies formulate implicit feedback as binary relevance feedback . However, in numerous applications, implicit feedback is observed not only as a binary indicator but also in a graded form, such as the number ...", "subpage_snippet": "", "source": "usait0.com", "link": "https://usait0.com/en/publication/workshops/wsdm2022/", "content": "Binary user-behavior logs such as clicks or views, called implicit feedback , are often used to build recommender systems because of its general availability in real practice. Most existing studies formulate implicit feedback as binary relevance feedback . However, in numerous applications, implicit feedback is observed not only as a binary indicator but also in a graded form, such as the number ..."}
+{"idx": 3, "title": "Dual Unbiased Recommender Learning for Implicit Feedback", "date": "", "ddg_snippet": "Abstract Unbiased recommender learning has been actively studied to alleviate the inherent bias of implicit datasets under the missing-not-at-random assumption. Existing studies solely address the bias of positive feedback but do not account for the bias of missing feedback , which heavily affects their sub-optimal performance gains.", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/10.1145/3404835.3463118", "content": "Abstract Unbiased recommender learning has been actively studied to alleviate the inherent bias of implicit datasets under the missing-not-at-random assumption. Existing studies solely address the bias of positive feedback but do not account for the bias of missing feedback , which heavily affects their sub-optimal performance gains."}
+{"idx": 4, "title": "Practically Unbiased Pairwise Loss for Recommendation With Implicit ...", "date": "", "ddg_snippet": "In this paper, we focus on unbiased ranking loss weighted by inversed propensity scores (IPS), which are widely used in recommendations with implicit feedback labels. More specifically, we first highlight the fact that there is a gap between theory and practice in IPS-weighted unbiased loss.", "subpage_snippet": "", "source": "pubmed.ncbi.nlm.nih.gov", "link": "https://pubmed.ncbi.nlm.nih.gov/40030662/", "content": "In this paper, we focus on unbiased ranking loss weighted by inversed propensity scores (IPS), which are widely used in recommendations with implicit feedback labels. More specifically, we first highlight the fact that there is a gap between theory and practice in IPS-weighted unbiased loss."}
+{"idx": 5, "title": "Unbiased Recommender Learning from Implicit Feedback via Weakly ...", "date": "", "ddg_snippet": "Poster Unbiased Recommender Learning from Implicit Feedback via Weakly Supervised Learning Eric Wang · Zhichao Chen · Haotian Wang · Yanchao Tan · Licheng Pan · Tianqiao Liu · Xu Chen · Haoxuan Li · Zhouchen Lin West Exhibition Hall B2-B3 #W-403", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/poster/46694", "content": "Poster Unbiased Recommender Learning from Implicit Feedback via Weakly Supervised Learning Eric Wang · Zhichao Chen · Haotian Wang · Yanchao Tan · Licheng Pan · Tianqiao Liu · Xu Chen · Haoxuan Li · Zhouchen Lin West Exhibition Hall B2-B3 #W-403"}
+{"idx": 6, "title": "Unbiased Recommender Learning from Missing-Not-At-Random Implicit Feedback", "date": "", "ddg_snippet": "Recommender systems widely use implicit feedback such as click data because of its general availability. Although the presence of clicks signals the users' preference to some extent, the lack of such clicks does not necessarily indicate a negative response from the users, as it is possible that the users were not exposed to the items (positive-unlabeled problem). This leads to a difficulty in ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/1909.03601", "content": "Recommender systems widely use implicit feedback such as click data because of its general availability. Although the presence of clicks signals the users' preference to some extent, the lack of such clicks does not necessarily indicate a negative response from the users, as it is possible that the users were not exposed to the items (positive-unlabeled problem). This leads to a difficulty in ..."}
+{"idx": 7, "title": "Unbiased Recommender Learning from Missing-Not-At-Random Implicit Feedback", "date": "", "ddg_snippet": "This repository accompanies the real-world experiment conducted in the paper \" Unbiased Recommender Learning from Missing-Not-At-Random Implicit Feedback \" by Yuta Saito, Suguru Yaginuma, Yuta Nishino, Hayato Sakata, and Kazuhide Nakata, which has been accepted to WSDM'20. If you find this code useful ...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/usaito/unbiased-implicit-rec-real", "content": "This repository accompanies the real-world experiment conducted in the paper \" Unbiased Recommender Learning from Missing-Not-At-Random Implicit Feedback \" by Yuta Saito, Suguru Yaginuma, Yuta Nishino, Hayato Sakata, and Kazuhide Nakata, which has been accepted to WSDM'20. If you find this code useful ..."}
+{"idx": 8, "title": "Unbiased Recommender Learning from Implicit Feedback via Weakly ...", "date": "", "ddg_snippet": "This paper formulates implicit feedback recommendation as a weakly supervised learning problem, obtaining an unbiased positive-negative recommender without the need of negative feedback .", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=0E5rZOGA13", "content": "This paper formulates implicit feedback recommendation as a weakly supervised learning problem, obtaining an unbiased positive-negative recommender without the need of negative feedback ."}
+{"idx": 9, "title": "arXiv:2304.05066v2 [cs.IR] 14 Apr 2023", "date": "", "ddg_snippet": "1 INTRODUCTION Recommender systems usually rely on implicit user feedback for model training owning to the cheap cost of collecting such data [17]. For this scenario, the typical model learning techniques [10, 14, 19], recognize interacted items as positive and all the other items as potential negative examples. There are mainly two dif-ficulties to learn unbiased user preference based on ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2304.05066", "content": "1 INTRODUCTION Recommender systems usually rely on implicit user feedback for model training owning to the cheap cost of collecting such data [17]. For this scenario, the typical model learning techniques [10, 14, 19], recognize interacted items as positive and all the other items as potential negative examples. There are mainly two dif-ficulties to learn unbiased user preference based on ..."}
diff --git a/data/sampled_jsons/Van_Stekelenburg_Klandermans_2013_collective_action_social_psychology_protest_year_2013.jsonl b/data/sampled_jsons/Van_Stekelenburg_Klandermans_2013_collective_action_social_psychology_protest_year_2013.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..52216e5778e80b6be9f67a8d9fbc15174dad2a04
--- /dev/null
+++ b/data/sampled_jsons/Van_Stekelenburg_Klandermans_2013_collective_action_social_psychology_protest_year_2013.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "(PDF) The Social Psychology of Protest", "date": "", "ddg_snippet": "Jacquelien van Stekelenburg and Bert Klandermans . The social psychology of protest . Collective action , emotions, grievances, identity, social psychology of protest . Why do people protest ?", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/258131447_The_Social_Psychology_of_Protest", "content": "Jacquelien van Stekelenburg and Bert Klandermans . The social psychology of protest . Collective action , emotions, grievances, identity, social psychology of protest . Why do people protest ?"}
+{"idx": 1, "title": "social movements and the dynamics of collective action Jacquelien...", "date": "", "ddg_snippet": "In The Social Psychology of Protest , Klandermans expands on this definition to distinguish among illegitimate inequality, suddenly imposed grievances, and violated principles.", "subpage_snippet": "", "source": "research.vu.nl", "link": "https://research.vu.nl/ws/portalfiles/portal/298859540/Ch._26_van_Stekelenburg_Gaidyte_social_movements_and_the_dynamics_of_collective_action.docx", "content": "In The Social Psychology of Protest , Klandermans expands on this definition to distinguish among illegitimate inequality, suddenly imposed grievances, and violated principles."}
+{"idx": 2, "title": "[PDF] The social psychology of protest | Semantic Scholar", "date": "", "ddg_snippet": "Jacquelien van Stekelenburg , B. Klandermans .ABSTRACT What drives a person to take part in a collective action and engage in political protest ? This is a question that has long interested social scientists. Recent theoretical and empirical…", "subpage_snippet": "", "source": "www.semanticscholar.org", "link": "https://www.semanticscholar.org/paper/The-social-psychology-of-protest-Stekelenburg-Klandermans/f6b747627f12725b9be8d1f1a070f8cf9d20c157", "content": "Jacquelien van Stekelenburg , B. Klandermans .ABSTRACT What drives a person to take part in a collective action and engage in political protest ? This is a question that has long interested social scientists. Recent theoretical and empirical…"}
+{"idx": 3, "title": "Ch 26 van Stekelenburg Gaidyte social movements and the dynamics...", "date": "", "ddg_snippet": "van Stekelenburg , Jacquelien, and Bert Klandermans . A Social Psychology of Protest : Individuals in Action . Cambridge University Press, 2024. In 1997 The Social Psychology of Protest ( Klandermans , 1997) appeared.", "subpage_snippet": "", "source": "www.academia.edu", "link": "https://www.academia.edu/115119854/Ch_26_van_Stekelenburg_Gaidyte_social_movements_and_the_dynamics_of_collective_action", "content": "van Stekelenburg , Jacquelien, and Bert Klandermans . A Social Psychology of Protest : Individuals in Action . Cambridge University Press, 2024. In 1997 The Social Psychology of Protest ( Klandermans , 1997) appeared."}
+{"idx": 4, "title": "Jacquelien van Stekelenburg - Google Scholar", "date": "", "ddg_snippet": "The social psychology of protest . J Van Stekelenburg , B Klandermans .J van Stekelenburg . The Wiley-Blackwell Encyclopedia of Social and Political Movements, 219-225, 2013 .", "subpage_snippet": "", "source": "scholar.google.com", "link": "https://scholar.google.com/citations?user=fP-OzuoAAAAJ&hl=de", "content": "The social psychology of protest . J Van Stekelenburg , B Klandermans .J van Stekelenburg . The Wiley-Blackwell Encyclopedia of Social and Political Movements, 219-225, 2013 ."}
+{"idx": 5, "title": "Collective Action as a Precursor to Social Change: Causes, Risks...", "date": "", "ddg_snippet": "In social psychology , collective action is often studied in the context of social movements, protests , and other collective political actions . Van Stekelenburg , J., & Klandermans , B. ( 2013 ). Social psychology of protest . Current Sociology, 61(5-6), 886-905.", "subpage_snippet": "", "source": "rsisinternational.org", "link": "https://rsisinternational.org/journals/ijriss/articles/collective-action-as-a-precursor-to-social-change-causes-risks-and-consequences/", "content": "In social psychology , collective action is often studied in the context of social movements, protests , and other collective political actions . Van Stekelenburg , J., & Klandermans , B. ( 2013 ). Social psychology of protest . Current Sociology, 61(5-6), 886-905."}
+{"idx": 6, "title": "Social Psychology Of Protest | Jacquelien van Stekelenburg Bert...", "date": "", "ddg_snippet": "Collective action is a matter of demand by citizens who are aggrieved, supply by protest organizations/individuals, and mobilization through effective communication networks. This volume elaborates on the processes and mechanisms responsible for these dynamics\"", "subpage_snippet": "", "source": "ru.z-lib.fm", "link": "https://ru.z-lib.fm/book/27053952/becfd4/social-psychology-of-protest.html?dsource=recommend", "content": "Collective action is a matter of demand by citizens who are aggrieved, supply by protest organizations/individuals, and mobilization through effective communication networks. This volume elaborates on the processes and mechanisms responsible for these dynamics\""}
+{"idx": 7, "title": "The Social Psychology of Protest - Bert Klandermans", "date": "", "ddg_snippet": "A Social Psychology of Protest : Individuals in Action Jacquelien van Stekelenburg ,Bert Klandermans Geen voorbeeld beschikbaar - 2023. Over de auteur (1997). Bert Klandermans is Professor of Applied Social Psychology at Free University, Amsterdam, The Netherlands.", "subpage_snippet": "", "source": "books.google.nl", "link": "https://books.google.nl/books/about/The_Social_Psychology_of_Protest.html?id=QId5QgAACAAJ&redir_esc=y", "content": "A Social Psychology of Protest : Individuals in Action Jacquelien van Stekelenburg ,Bert Klandermans Geen voorbeeld beschikbaar - 2023. Over de auteur (1997). Bert Klandermans is Professor of Applied Social Psychology at Free University, Amsterdam, The Netherlands."}
+{"idx": 8, "title": "We are all in this together: Psychology of Social Movement | Medium", "date": "", "ddg_snippet": "The social psychology of protest . Current Sociology, 61(5–6), 886–90. Van Zomeren, M., Postmes, T., & Spears, R. (2008). Toward an integrative social identity model of collective action : a quantitative research synthesis of three socio- psychological perspectives.", "subpage_snippet": "", "source": "medium.com", "link": "https://medium.com/@sreyadas0701/we-are-all-in-this-together-psychology-of-social-movement-c2be94ed4b77", "content": "The social psychology of protest . Current Sociology, 61(5–6), 886–90. Van Zomeren, M., Postmes, T., & Spears, R. (2008). Toward an integrative social identity model of collective action : a quantitative research synthesis of three socio- psychological perspectives."}
+{"idx": 9, "title": "Why nonviolent protests work better - Tue, October... - The Jakarta Post", "date": "", "ddg_snippet": "Research from Stekelenburg and Klandermans ( 2013 ) in The Social Psychology of Protest , suggests that grievance and efficacy would predict protest participation.", "subpage_snippet": "", "source": "www.thejakartapost.com", "link": "https://www.thejakartapost.com/paper/2020/10/26/why-nonviolent-protests-work-better.html", "content": "Research from Stekelenburg and Klandermans ( 2013 ) in The Social Psychology of Protest , suggests that grievance and efficacy would predict protest participation."}
diff --git a/data/sampled_jsons/Video-ColBERT_MSVD_frames_sampled_Section_5.2_siteopenaccess.thecvf.com_year_2023.jsonl b/data/sampled_jsons/Video-ColBERT_MSVD_frames_sampled_Section_5.2_siteopenaccess.thecvf.com_year_2023.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..c718294efad9cf4ef294aab01e2aaaa2ac2d7906
--- /dev/null
+++ b/data/sampled_jsons/Video-ColBERT_MSVD_frames_sampled_Section_5.2_siteopenaccess.thecvf.com_year_2023.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "PDF Video-ColBERT: Contextualized Late Interaction for Text-to-Video Retrieval", "date": "", "ddg_snippet": "Abstract In this work, we tackle the problem of text-to- video re-trieval (T2VR). Inspired by the success of late interaction techniques in text-document, text-image, and text- video re-trieval, our approach, Video-ColBERT , introduces a simple and eficient mechanism for fine-grained similarity assess-ment between queries and videos .", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content/CVPR2025/papers/Reddy_Video-ColBERT_Contextualized_Late_Interaction_for_Text-to-Video_Retrieval_CVPR_2025_paper.pdf", "content": "Abstract In this work, we tackle the problem of text-to- video re-trieval (T2VR). Inspired by the success of late interaction techniques in text-document, text-image, and text- video re-trieval, our approach, Video-ColBERT , introduces a simple and eficient mechanism for fine-grained similarity assess-ment between queries and videos ."}
+{"idx": 1, "title": "HAIR: Hierarchical Visual-Semantic Relational Reasoning for Video ...", "date": "", "ddg_snippet": "We evenly sample 10 frames to represent the video and select 6 detected objects with the highest scores per frame . The dimensionality of the joint embedding space d is 512. The number of visual and se-mantic reasoning steps, Kv and Ks, are set to 2 and 2, re-spectively.", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content/ICCV2021/papers/Liu_HAIR_Hierarchical_Visual-Semantic_Relational_Reasoning_for_Video_Question_Answering_ICCV_2021_paper.pdf", "content": "We evenly sample 10 frames to represent the video and select 6 detected objects with the highest scores per frame . The dimensionality of the joint embedding space d is 512. The number of visual and se-mantic reasoning steps, Kv and Ks, are set to 2 and 2, re-spectively."}
+{"idx": 2, "title": "PDF Video-ColBERT: Contextualized Late Interaction for Text-to-Video ...", "date": "", "ddg_snippet": "Visualization of the interactions between query tokens and video frames before and after the temporal encoder of VIDEO - COLBERT , trained on MSR-VTT. The green arrow () represents the interaction between query tokens and frames before temporal encod- ing.", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content/CVPR2025/supplemental/Reddy_Video-ColBERT_Contextualized_Late_CVPR_2025_supplemental.pdf", "content": "Visualization of the interactions between query tokens and video frames before and after the temporal encoder of VIDEO - COLBERT , trained on MSR-VTT. The green arrow () represents the interaction between query tokens and frames before temporal encod- ing."}
+{"idx": 3, "title": "PDF ICSVR: Investigating Compositional and Syntactic Understanding in Video ...", "date": "", "ddg_snippet": "The table shows the results on MSVD [4] dataset in both text-to- video and video -to-text retrieval settings. Q denotes the performance (R@1 score) on the original unchanged dataset.", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content/CVPR2024W/MMFM/papers/Madasu_ICSVR_Investigating_Compositional_and_Syntactic_Understanding_in_Video_Retrieval_Models_CVPRW_2024_paper.pdf", "content": "The table shows the results on MSVD [4] dataset in both text-to- video and video -to-text retrieval settings. Q denotes the performance (R@1 score) on the original unchanged dataset."}
+{"idx": 4, "title": "PDF MV-Adapter: Multimodal Video Transfer Learning for Video Text Retrieval", "date": "", "ddg_snippet": "First, we introduce a temporal adaptation (TA) mod-ule in the video branch to enhance the temporal modeling capability. Unlike previous video adapters that apply iden-tical weights across frames , we generate dynamic weights from both global and local features to better capture tempo-ral variations in videos .", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content/CVPR2024/papers/Jin_MV-Adapter_Multimodal_Video_Transfer_Learning_for_Video_Text_Retrieval_CVPR_2024_paper.pdf", "content": "First, we introduce a temporal adaptation (TA) mod-ule in the video branch to enhance the temporal modeling capability. Unlike previous video adapters that apply iden-tical weights across frames , we generate dynamic weights from both global and local features to better capture tempo-ral variations in videos ."}
+{"idx": 5, "title": "PDF MSR-VTT: A Large Video Description Dataset for Bridging Video and Language", "date": "", "ddg_snippet": "In this paper we present MSR-VTT (standing for \"MSR- Video to Text) which is a new large-scale video bench-mark for video understanding, especially the emerging task of translating video to text. This is achieved by collecting 257 popular queries from a commercial video search en-gine, with 118 videos for each query. In its current ver-sion, MSR-VTT provides 10K web video clips with 41.2 ...", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content_cvpr_2016/papers/Xu_MSR-VTT_A_Large_CVPR_2016_paper.pdf", "content": "In this paper we present MSR-VTT (standing for \"MSR- Video to Text) which is a new large-scale video bench-mark for video understanding, especially the emerging task of translating video to text. This is achieved by collecting 257 popular queries from a commercial video search en-gine, with 118 videos for each query. In its current ver-sion, MSR-VTT provides 10K web video clips with 41.2 ..."}
+{"idx": 6, "title": "PDF Less Is More: Picking Informative Frames for Video Captioning", "date": "", "ddg_snippet": "It is also recognized as the word- frame association learned by sparse coding [41] or gaze-guided attention learning [45], which is a de-facto frame weighting mechanism. This mechanism also benefits many downstream tasks such as visual captioning and visual question answer-ing for image and video [20, 43, 12].", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content_ECCV_2018/papers/Yangyu_Chen_Less_is_More_ECCV_2018_paper.pdf", "content": "It is also recognized as the word- frame association learned by sparse coding [41] or gaze-guided attention learning [45], which is a de-facto frame weighting mechanism. This mechanism also benefits many downstream tasks such as visual captioning and visual question answer-ing for image and video [20, 43, 12]."}
+{"idx": 7, "title": "PDF Video Captioning With Transferred Semantic Attributes - CVF Open Access", "date": "", "ddg_snippet": "The results across different metrics consis-tently indicate that LSTM-TSAV with semantic attributes learnt by video MIL model leads to a better performance, demonstrating the advantage of exploring semantic infor-mation among all the sampled frames from one video holis-tically, as opposed to locally based on individual frame .", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content_cvpr_2017/papers/Pan_Video_Captioning_With_CVPR_2017_paper.pdf", "content": "The results across different metrics consis-tently indicate that LSTM-TSAV with semantic attributes learnt by video MIL model leads to a better performance, demonstrating the advantage of exploring semantic infor-mation among all the sampled frames from one video holis-tically, as opposed to locally based on individual frame ."}
+{"idx": 8, "title": "PDF Weakly Supervised Video Representation Learning with Unaligned Text for ...", "date": "", "ddg_snippet": "In the vision representation module, we feed the frames sampled from the untrimmed sequential video into the module, then obtain the frame representations and a video representation.", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content/CVPR2023/papers/Dong_Weakly_Supervised_Video_Representation_Learning_With_Unaligned_Text_for_Sequential_CVPR_2023_paper.pdf", "content": "In the vision representation module, we feed the frames sampled from the untrimmed sequential video into the module, then obtain the frame representations and a video representation."}
+{"idx": 9, "title": "PDF MELTR: Meta Loss Transformer for Learning to Fine-tune Video Foundation ...", "date": "", "ddg_snippet": "When only the SE is fed in Figure 1(a), MELTR tends to focus on reasonably chal-lenging samples and downweight the noisy samples as dis-cussed in Section 5.2 of the main paper.", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content/CVPR2023/supplemental/Ko_MELTR_Meta_Loss_CVPR_2023_supplemental.pdf", "content": "When only the SE is fed in Figure 1(a), MELTR tends to focus on reasonably chal-lenging samples and downweight the noisy samples as dis-cussed in Section 5.2 of the main paper."}
diff --git a/data/sampled_jsons/Waymo_Open_Dataset_Sun_et_al._2020_1150_scenes_20_seconds.jsonl b/data/sampled_jsons/Waymo_Open_Dataset_Sun_et_al._2020_1150_scenes_20_seconds.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..3da83fa9ece8d38a7868568adc66bdf434fe4b90
--- /dev/null
+++ b/data/sampled_jsons/Waymo_Open_Dataset_Sun_et_al._2020_1150_scenes_20_seconds.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Scalability in Perception for Autonomous Driving: Waymo Open Dataset", "date": "", "ddg_snippet": "Our new dataset consists of 1150 scenes that each span 20 seconds , consisting of well synchronized and calibrated high quality LiDAR and camera data captured across a range of urban and suburban geographies. It is 15x more diverse than the largest camera+LiDAR dataset available based on our proposed diversity metric.", "subpage_snippet": "", "source": "waymo.com", "link": "https://waymo.com/research/scalability-in-perception-for-autonomous-driving-waymo-open-dataset/", "content": "Our new dataset consists of 1150 scenes that each span 20 seconds , consisting of well synchronized and calibrated high quality LiDAR and camera data captured across a range of urban and suburban geographies. It is 15x more diverse than the largest camera+LiDAR dataset available based on our proposed diversity metric."}
+{"idx": 1, "title": "CVPR 2020 Open Access Repository", "date": "", "ddg_snippet": "Our new dataset consists of 1150 scenes that each span 20 seconds , consisting of well synchronized and calibrated high quality LiDAR and camera data captured across a range of urban and suburban geographies. It is 15x more diverse than the largest camera+LiDAR dataset available based on our proposed diversity metric.", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content_CVPR_2020/html/Sun_Scalability_in_Perception_for_Autonomous_Driving_Waymo_Open_Dataset_CVPR_2020_paper.html", "content": "Our new dataset consists of 1150 scenes that each span 20 seconds , consisting of well synchronized and calibrated high quality LiDAR and camera data captured across a range of urban and suburban geographies. It is 15x more diverse than the largest camera+LiDAR dataset available based on our proposed diversity metric."}
+{"idx": 2, "title": "waymo-research/waymo-open-dataset - GitHub", "date": "", "ddg_snippet": "The Waymo Open Dataset is a collection of datasets and evaluation code that we have released publicly to aid the research community in making advancements in machine perception and autonomous driving technology. The Waymo Open Dataset includes three datasets : The Perception dataset , with high resolution sensor data and labels for various tasks.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/waymo-research/waymo-open-dataset", "content": "The Waymo Open Dataset is a collection of datasets and evaluation code that we have released publicly to aid the research community in making advancements in machine perception and autonomous driving technology. The Waymo Open Dataset includes three datasets : The Perception dataset , with high resolution sensor data and labels for various tasks."}
+{"idx": 3, "title": "Scalability in Perception for Autonomous Driving: Waymo Open Dataset", "date": "", "ddg_snippet": "Our new dataset consists of 1150 scenes that each span 20 seconds , consisting of well synchronized and calibrated high quality LiDAR and camera data captured across a range of urban and suburban geographies. It is 15x more diverse than the largest camera+LiDAR dataset available based on our proposed diversity metric.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/1912.04838", "content": "Our new dataset consists of 1150 scenes that each span 20 seconds , consisting of well synchronized and calibrated high quality LiDAR and camera data captured across a range of urban and suburban geographies. It is 15x more diverse than the largest camera+LiDAR dataset available based on our proposed diversity metric."}
+{"idx": 4, "title": "PDF Waymo Open Dataset: Panoramic Video Panoptic Segmentation - ECVA", "date": "", "ddg_snippet": "The Waymo Open Dataset contains 1,150 scenes , each consisting of 20 seconds of data captured at 10Hz (i.e., 10 frames per second , and thus 200 frames per scene ).", "subpage_snippet": "", "source": "www.ecva.net", "link": "https://www.ecva.net/papers/eccv_2022/papers_ECCV/papers/136890052.pdf", "content": "The Waymo Open Dataset contains 1,150 scenes , each consisting of 20 seconds of data captured at 10Hz (i.e., 10 frames per second , and thus 200 frames per scene )."}
+{"idx": 5, "title": "Processing, assessing, and enhancing the Waymo autonomous vehicle open ...", "date": "", "ddg_snippet": "The Waymo Open Dataset ( Sun et al ., 2020b) enables such detailed assessment as it includes fineresolution motion information of AVs in mixed autonomy traffic from 1,000 segments (2019 release) at ...", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/356748756_Processing_assessing_and_enhancing_the_Waymo_autonomous_vehicle_open_dataset_for_driving_behavior_research", "content": "The Waymo Open Dataset ( Sun et al ., 2020b) enables such detailed assessment as it includes fineresolution motion information of AVs in mixed autonomy traffic from 1,000 segments (2019 release) at ..."}
+{"idx": 6, "title": "The Waymo Open Dataset | av-dataset - Weights & Biases", "date": "", "ddg_snippet": "The perception dataset consists of 1150 scenes that each span 20 seconds , consisting of well synchronized and calibrated high-quality LiDAR and camera data captured across a range of urban and suburban geographies. This includes high-resolution sensor data and labels for 2,030 segments, key points labels, 2D-to-3D association labels, 3D semantic segmentation labels, and 2D video panoptic ...", "subpage_snippet": "", "source": "wandb.ai", "link": "https://wandb.ai/av-datasets/av-dataset/reports/The-Waymo-Open-Dataset--VmlldzoyNjI0NTYy", "content": "The perception dataset consists of 1150 scenes that each span 20 seconds , consisting of well synchronized and calibrated high-quality LiDAR and camera data captured across a range of urban and suburban geographies. This includes high-resolution sensor data and labels for 2,030 segments, key points labels, 2D-to-3D association labels, 3D semantic segmentation labels, and 2D video panoptic ..."}
+{"idx": 7, "title": "Scalability in Perception for Autonomous Driving: Waymo Open Dataset ...", "date": "", "ddg_snippet": "The research community has increasing interest in autonomous driving research, despite the resource intensity of obtaining representative real world data. Existing self-driving datasets are limited in the scale and variation of the environments they capture, even though generalization within and between operating regions is crucial to the over-all viability of the technology. In an effort to ...", "subpage_snippet": "", "source": "ieeexplore.ieee.org", "link": "https://ieeexplore.ieee.org/document/9156973", "content": "The research community has increasing interest in autonomous driving research, despite the resource intensity of obtaining representative real world data. Existing self-driving datasets are limited in the scale and variation of the environments they capture, even though generalization within and between operating regions is crucial to the over-all viability of the technology. In an effort to ..."}
+{"idx": 8, "title": "Waymo Open dataset - Dataset - LDM - service.tib.eu", "date": "", "ddg_snippet": "The Waymo Open dataset provides 798 training sequences and 202 validation sequences. Each sequence contains 200 frames within 20 seconds .", "subpage_snippet": "", "source": "service.tib.eu", "link": "https://service.tib.eu/ldmservice/dataset/waymo-open-dataset", "content": "The Waymo Open dataset provides 798 training sequences and 202 validation sequences. Each sequence contains 200 frames within 20 seconds ."}
+{"idx": 9, "title": "PDF Scalability in Perception for Autonomous Driving: Waymo Open Dataset", "date": "", "ddg_snippet": "Our new dataset consists of 1150 scenes that each span 20 seconds , consisting of well syn-chronized and calibrated high quality LiDAR and camera data captured across a range of urban and suburban ge-ographies. It is 15x more diverse than the largest cam-era+LiDAR dataset available based on our proposed geo-graphical coverage metric.", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content_CVPR_2020/papers/Sun_Scalability_in_Perception_for_Autonomous_Driving_Waymo_Open_Dataset_CVPR_2020_paper.pdf", "content": "Our new dataset consists of 1150 scenes that each span 20 seconds , consisting of well syn-chronized and calibrated high quality LiDAR and camera data captured across a range of urban and suburban ge-ographies. It is 15x more diverse than the largest cam-era+LiDAR dataset available based on our proposed geo-graphical coverage metric."}
diff --git a/data/sampled_jsons/When_building_a_unified_vision_system_Learning_without_Forgetting_Li_Hoiem.jsonl b/data/sampled_jsons/When_building_a_unified_vision_system_Learning_without_Forgetting_Li_Hoiem.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..78682705d0edc7b15ea2192a6d08a358d49665c7
--- /dev/null
+++ b/data/sampled_jsons/When_building_a_unified_vision_system_Learning_without_Forgetting_Li_Hoiem.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "[1606.09282] Learning without Forgetting", "date": "", "ddg_snippet": "by Z Li · 2016 · Cited by 5971 — Title: Learning without Forgetting ... Abstract: When building a unified vision system or gradually adding new capabilities to a system, the usual ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/1606.09282", "content": "by Z Li · 2016 · Cited by 5971 — Title: Learning without Forgetting ... Abstract: When building a unified vision system or gradually adding new capabilities to a system, the usual ..."}
+{"idx": 1, "title": "Learning without Forgetting", "date": "", "ddg_snippet": "by Z Li · Cited by 5956 — Abstract— When building a unified vision system or gradually adding new ... Hoiem , “ Learning without forgetting ,” in European. Conference on Computer Vision. 13 pages", "subpage_snippet": "", "source": "ieeexplore.ieee.org", "link": "https://ieeexplore.ieee.org/ielaam/34/8520726/8107520-aam.pdf", "content": "by Z Li · Cited by 5956 — Abstract— When building a unified vision system or gradually adding new ... Hoiem , “ Learning without forgetting ,” in European. Conference on Computer Vision. 13 pages"}
+{"idx": 2, "title": "Learning without forgetting", "date": "", "ddg_snippet": "When building a unified vision system or gradually adding new capabilities ... In brief, our Learning without Forgetting approach could be seen as a combi-. 16 pages", "subpage_snippet": "", "source": "scispace.com", "link": "https://scispace.com/pdf/learning-without-forgetting-34a4300f5h.pdf", "content": "When building a unified vision system or gradually adding new capabilities ... In brief, our Learning without Forgetting approach could be seen as a combi-. 16 pages"}
+{"idx": 3, "title": "Learning without Forgetting", "date": "", "ddg_snippet": "by Z Li · 2016 · Cited by 5959 — Abstract— When building a unified vision system or ... Hoiem , “ Learning without forgetting ,” in European. Conference on Computer Vision.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/1606.09282", "content": "by Z Li · 2016 · Cited by 5959 — Abstract— When building a unified vision system or ... Hoiem , “ Learning without forgetting ,” in European. Conference on Computer Vision."}
+{"idx": 4, "title": "Learning without forgetting", "date": "", "ddg_snippet": "by Z Li · 2016 · Cited by 5959 — When building a unified vision system or gradually adding new ... Li , Z & Hoiem , D 2016, Learning without forgetting . in B Leibe, J Matas, N ...", "subpage_snippet": "", "source": "experts.illinois.edu", "link": "https://experts.illinois.edu/en/publications/learning-without-forgetting", "content": "by Z Li · 2016 · Cited by 5959 — When building a unified vision system or gradually adding new ... Li , Z & Hoiem , D 2016, Learning without forgetting . in B Leibe, J Matas, N ..."}
+{"idx": 5, "title": "Learning without Forgetting | IEEE Transactions on Pattern ...", "date": "", "ddg_snippet": "When building a unified vision system or gradually adding new apabilities to ... Hoiem , “ Learning without forgetting ,” in Proc. Eur. Conf. Comput. Vis ...", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/10.1109/tpami.2017.2773081", "content": "When building a unified vision system or gradually adding new apabilities to ... Hoiem , “ Learning without forgetting ,” in Proc. Eur. Conf. Comput. Vis ..."}
+{"idx": 6, "title": "Learning Without Forgetting: Zhizhong Li, Derek Hoiem, Member ...", "date": "", "ddg_snippet": "Learning without Forgetting . Zhizhong Li , Derek Hoiem , Member, IEEE. Abstract— When building a unified vision system or gradually adding new capabilities to a ...", "subpage_snippet": "", "source": "www.scribd.com", "link": "https://www.scribd.com/document/763628484/1606-09282", "content": "Learning without Forgetting . Zhizhong Li , Derek Hoiem , Member, IEEE. Abstract— When building a unified vision system or gradually adding new capabilities to a ..."}
+{"idx": 7, "title": "(PDF) Learning without Forgetting (2018) | Zhizhong Li", "date": "", "ddg_snippet": "... (DOI: 10.1109/TPAMI.2017.2773081) When building a unified vision system or ... Learning without Forgetting method, which uses only new task data to ...", "subpage_snippet": "", "source": "scispace.com", "link": "https://scispace.com/papers/learning-without-forgetting-1csfq4yxg5", "content": "... (DOI: 10.1109/TPAMI.2017.2773081) When building a unified vision system or ... Learning without Forgetting method, which uses only new task data to ..."}
+{"idx": 8, "title": "Learning without Forgetting", "date": "", "ddg_snippet": "Paper. Cite Save. Learning without Forgetting . Published Jun 29, 2016 · Zhizhong Li , Derek Hoiem ... When building a unified vision system or gradually adding ...", "subpage_snippet": "", "source": "consensus.app", "link": "https://consensus.app/papers/learning-without-forgetting-li-hoiem/37e70e6c613d5dffba39c61780398bb6/", "content": "Paper. Cite Save. Learning without Forgetting . Published Jun 29, 2016 · Zhizhong Li , Derek Hoiem ... When building a unified vision system or gradually adding ..."}
+{"idx": 9, "title": "Learning without Forgetting", "date": "", "ddg_snippet": "by Z Li · 2017 · Cited by 5959 — Abstract— When building a unified vision system or gradually adding new apabilities to a ... LI AND HOIEM : LEARNING WITHOUT FORGETTING . 2947. 13 pages", "subpage_snippet": "", "source": "ieeexplore.ieee.org", "link": "https://ieeexplore.ieee.org/iel7/34/8520726/08107520.pdf", "content": "by Z Li · 2017 · Cited by 5959 — Abstract— When building a unified vision system or gradually adding new apabilities to a ... LI AND HOIEM : LEARNING WITHOUT FORGETTING . 2947. 13 pages"}
diff --git a/data/sampled_jsons/WiSE-FT_ensembling_the_weights_of_the_zero-shot_and_fine-tuned_models_abstract_year_2021.jsonl b/data/sampled_jsons/WiSE-FT_ensembling_the_weights_of_the_zero-shot_and_fine-tuned_models_abstract_year_2021.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..d9628d454feb6366866ad579bb9874d42945d781
--- /dev/null
+++ b/data/sampled_jsons/WiSE-FT_ensembling_the_weights_of_the_zero-shot_and_fine-tuned_models_abstract_year_2021.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "[2109.01903] Robust fine-tuning of zero-shot models", "date": "", "ddg_snippet": "... a simple and effective method for improving robustness while fine -tuning: ensembling the weights of the zero - shot and fine - tuned models ( WiSE - FT ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2109.01903", "content": "... a simple and effective method for improving robustness while fine -tuning: ensembling the weights of the zero - shot and fine - tuned models ( WiSE - FT ..."}
+{"idx": 1, "title": "Fine-Tuning Language Models with Just Forward Passes", "date": "", "ddg_snippet": "... 13B results with zero - shot , in-context learning (ICL), MeZO (we report the best among MeZO/MeZO (LoRA)/MeZO (prefix)), and fine -tuning with Adam ( FT ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2305.17333v3", "content": "... 13B results with zero - shot , in-context learning (ICL), MeZO (we report the best among MeZO/MeZO (LoRA)/MeZO (prefix)), and fine -tuning with Adam ( FT ..."}
+{"idx": 2, "title": "Model Stock: All we need is just a few fine-tuned models", "date": "", "ddg_snippet": "Drawing from key insights in the weight space of fine - tuned weights , we uncover a strong link between the performance and proximity to the center of ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2403.19522v1", "content": "Drawing from key insights in the weight space of fine - tuned weights , we uncover a strong link between the performance and proximity to the center of ..."}
+{"idx": 3, "title": "WiSE-OD: Benchmarking Robustness in Infrared Object Detection", "date": "", "ddg_snippet": "22 ] proposed WiSE - FT , which ensembles the weights of a zero - shot pretrained model and its fine - tuned counterpart in weight space, showing strong ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2507.18925v1", "content": "22 ] proposed WiSE - FT , which ensembles the weights of a zero - shot pretrained model and its fine - tuned counterpart in weight space, showing strong ..."}
+{"idx": 4, "title": "Physics-Constrained Fine-Tuning of Flow-Matching Models for", "date": "", "ddg_snippet": "... and oceanographic modelling, seismic inversion, and medical imaging, we often observe system states without access to the underlying physical ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2508.09156v1", "content": "... and oceanographic modelling, seismic inversion, and medical imaging, we often observe system states without access to the underlying physical ..."}
+{"idx": 5, "title": "Prompt Tuning Vision Language Models with Margin Regularizer", "date": "", "ddg_snippet": "... the pre-training data of such models are unavailable, it is difficult to comprehend the performance on various downstream datasets.First, we try to ...", "subpage_snippet": "", "source": "parentscouncilofnashville.org", "link": "https://parentscouncilofnashville.org/article/prompt-tuning-vision-language-models-with-margin-regularizer-for-few-shot-learning-under-distribution-shifts", "content": "... the pre-training data of such models are unavailable, it is difficult to comprehend the performance on various downstream datasets.First, we try to ..."}
+{"idx": 6, "title": "Hung-yi LEE | Phd | National Taiwan University, Taipei | NTU |", "date": "", "ddg_snippet": "... each time someone views a publication summary (such as the title, abstract , and list of authors), clicks on a figure, or views or downloads the full ...", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/profile/Hung-Yi-Lee-2", "content": "... each time someone views a publication summary (such as the title, abstract , and list of authors), clicks on a figure, or views or downloads the full ..."}
+{"idx": 7, "title": "Tag: #TODO", "date": "", "ddg_snippet": "A detailed look at the state of Flood Forecasting in climate science and the integration of Machine Learning Pipelines", "subpage_snippet": "", "source": "stevengong.co", "link": "https://stevengong.co/tags/TODO", "content": "A detailed look at the state of Flood Forecasting in climate science and the integration of Machine Learning Pipelines"}
+{"idx": 8, "title": "Science Cast", "date": "", "ddg_snippet": "Abstract : In this work, we present CleanUNet 2, a speech denoising model that combines the advantages of waveform denoiser and spectrogram denoiser ...", "subpage_snippet": "", "source": "sciencecast.org", "link": "https://sciencecast.org/podcasts/arxiv_daily/episodes/3384", "content": "Abstract : In this work, we present CleanUNet 2, a speech denoising model that combines the advantages of waveform denoiser and spectrogram denoiser ..."}
+{"idx": 9, "title": "Prof. Dr. Bastian Leibe - Computer Vision", "date": "", "ddg_snippet": "... fine -tuning on top of the single-step model with task-specific losses and get a deterministic model that outperforms all other diffusion- based depth ...", "subpage_snippet": "", "source": "web-info8.informatik.rwth-aachen.de", "link": "http://web-info8.informatik.rwth-aachen.de/person/1/", "content": "... fine -tuning on top of the single-step model with task-specific losses and get a deterministic model that outperforms all other diffusion- based depth ..."}
diff --git a/data/sampled_jsons/Wortsman_et_al._2021_WiSE-FT_year_2021.jsonl b/data/sampled_jsons/Wortsman_et_al._2021_WiSE-FT_year_2021.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..0cc74db242810617df9cce32d937829b4664eb2e
--- /dev/null
+++ b/data/sampled_jsons/Wortsman_et_al._2021_WiSE-FT_year_2021.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "mlfoundations/ wise - ft - Githubissues", "date": "", "ddg_snippet": "WiSE - FT can be implemented in a few lines of code in addition to standard fine-tuning, as shown below. See src/ wise _ ft .py for more details.", "subpage_snippet": "", "source": "githubissues.com", "link": "https://githubissues.com/mlfoundations/wise-ft/readme", "content": "WiSE - FT can be implemented in a few lines of code in addition to standard fine-tuning, as shown below. See src/ wise _ ft .py for more details."}
+{"idx": 1, "title": "Figure 4: FLOW is complementary with model averaging ( WiSE - FT ) in...", "date": "", "ddg_snippet": "We compare WiSE - FT [ Wortsman et al ., 2021 ] with a standard model fine-tuning and with FLOW after fine-tuning Gemma 2 2B on MetaMathQA.The results indicate that combining Wise - FT with FLOW outperforms vanilla WiSE - FT with standard fine-tuning.", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/figure/FLOW-is-complementary-with-model-averaging-WiSE-FT-in-language-modeling-We-compare_fig2_388754394", "content": "We compare WiSE - FT [ Wortsman et al ., 2021 ] with a standard model fine-tuning and with FLOW after fine-tuning Gemma 2 2B on MetaMathQA.The results indicate that combining Wise - FT with FLOW outperforms vanilla WiSE - FT with standard fine-tuning."}
+{"idx": 2, "title": "Upweighting Easy Samples in Fine-Tuning Mitigates Forgetting", "date": "", "ddg_snippet": "We also consider model averaging as a baseline, specifically focusing on WiSE - FT ( Wortsman et al ., 2021 ) . WiSE - FT is simply the convex combination of the model parameters shared between the two tasks, while the task-specific parts are not averaged.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2502.02797v2", "content": "We also consider model averaging as a baseline, specifically focusing on WiSE - FT ( Wortsman et al ., 2021 ) . WiSE - FT is simply the convex combination of the model parameters shared between the two tasks, while the task-specific parts are not averaged."}
+{"idx": 3, "title": "Proceedings of the International Conference on Machine Learning 2022", "date": "", "ddg_snippet": "Wortsman et al . ( 2021 ) introduce WiSE - FT , a method for improving the robustness of a model θ1 which is fine-tuned from initialization θ0 by linearly interpolating θ1 and θ0.", "subpage_snippet": "", "source": "proceedings.mlr.press", "link": "https://proceedings.mlr.press/v162/wortsman22a/wortsman22a.pdf", "content": "Wortsman et al . ( 2021 ) introduce WiSE - FT , a method for improving the robustness of a model θ1 which is fine-tuned from initialization θ0 by linearly interpolating θ1 and θ0."}
+{"idx": 4, "title": "F orgetting and e nhances M odel M erging", "date": "", "ddg_snippet": "WiSE - FT 64. WImiaSgEe-NFeTtR+ LiNeS. Fine-tuned 67.We first consider the setting of robust fine-tuning or WiSE - FT ( Wortsman et al ., 2022b), where lin-early interpolating between the pre-trained and the fine-tuned weights improves model performance on OOD datasets.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=J5sUOvlLbQ", "content": "WiSE - FT 64. WImiaSgEe-NFeTtR+ LiNeS. Fine-tuned 67.We first consider the setting of robust fine-tuning or WiSE - FT ( Wortsman et al ., 2022b), where lin-early interpolating between the pre-trained and the fine-tuned weights improves model performance on OOD datasets."}
+{"idx": 5, "title": "Unleashing the Potentials of Likelihood Composition for Multi-modal", "date": "", "ddg_snippet": "Many works focus on getting a new model by inheriting the knowledge from multiple parent mod-els. Some of them interpolate several models’ weights to get the new model’s weight, e.g., WiSE - FT ( Wortsman et al ., 2021 ) and model soup ( Worts - man et al ., 2022).", "subpage_snippet": "", "source": "aclanthology.org", "link": "https://aclanthology.org/2024.findings-emnlp.594.pdf", "content": "Many works focus on getting a new model by inheriting the knowledge from multiple parent mod-els. Some of them interpolate several models’ weights to get the new model’s weight, e.g., WiSE - FT ( Wortsman et al ., 2021 ) and model soup ( Worts - man et al ., 2022)."}
+{"idx": 6, "title": "Finetune Like You Pretrain: Improved Finetuning of Zero-Shot Vision...", "date": "", "ddg_snippet": "Weight ensembling curves: WiSE - FT ( Wortsman et al ., 2021 ) shows that a simple linear interpolation between the weights of the pretrained and the finetuned model gives the best of both ID and OOD performance.", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content/CVPR2023/papers/Goyal_Finetune_Like_You_Pretrain_Improved_Finetuning_of_Zero-Shot_Vision_Models_CVPR_2023_paper.pdf", "content": "Weight ensembling curves: WiSE - FT ( Wortsman et al ., 2021 ) shows that a simple linear interpolation between the weights of the pretrained and the finetuned model gives the best of both ID and OOD performance."}
+{"idx": 7, "title": "GitHub - szubing/uniood: A code framework for implementing UniDA", "date": "", "ddg_snippet": "Robust fine-tuning of zero-shot models ( WiSE - FT , Wortsman et al ., 2022). Multimodality helps unimodality: Cross-modal few-shot learning with multimodal models (CLIP cross-model, Lin et al ., 2023).", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/szubing/uniood", "content": "Robust fine-tuning of zero-shot models ( WiSE - FT , Wortsman et al ., 2022). Multimodality helps unimodality: Cross-modal few-shot learning with multimodal models (CLIP cross-model, Lin et al ., 2023)."}
+{"idx": 8, "title": "Merging Models with Fisher-Weighted Averaging", "date": "", "ddg_snippet": "Recently, Wortsman et al . [66] demonstrated that merging can be used to improve robustness to domain shift in ne-tuned models by averaging the parameters of the original pre-trained model with the ne-tuned parameters.", "subpage_snippet": "", "source": "proceedings.neurips.cc", "link": "https://proceedings.neurips.cc/paper_files/paper/2022/file/70c26937fbf3d4600b69a129031b66ec-Paper-Conference.pdf", "content": "Recently, Wortsman et al . [66] demonstrated that merging can be used to improve robustness to domain shift in ne-tuned models by averaging the parameters of the original pre-trained model with the ne-tuned parameters."}
+{"idx": 9, "title": "Do the Frankenstein, or how to achieve better out-of-distribution", "date": "", "ddg_snippet": "The authors of [ Wortsman et al ., 2022b] found that while fine-tuning a pretrained vision model improves performance on the downstream task, it also tends to decrease accuracy on the original pretraining task.", "subpage_snippet": "", "source": "scispace.com", "link": "https://scispace.com/pdf/do-the-frankenstein-or-how-to-achieve-better-out-of-5d6q8x4mnf.pdf", "content": "The authors of [ Wortsman et al ., 2022b] found that while fine-tuning a pretrained vision model improves performance on the downstream task, it also tends to decrease accuracy on the original pretraining task."}
diff --git a/data/sampled_jsons/X-CLIP_End-to-End_Multi-grained_Contrastive_Learning_for_Video-Text_Retrieval_abstract_year_2022.jsonl b/data/sampled_jsons/X-CLIP_End-to-End_Multi-grained_Contrastive_Learning_for_Video-Text_Retrieval_abstract_year_2022.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..e61ac49811bc13329867ca6f161162d5791026db
--- /dev/null
+++ b/data/sampled_jsons/X-CLIP_End-to-End_Multi-grained_Contrastive_Learning_for_Video-Text_Retrieval_abstract_year_2022.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "[2207.07285] X-CLIP: End-to-End Multi-grained Contrastive ...", "date": "", "ddg_snippet": "Jul 15, 2022 · View a PDF of the paper titled X-CLIP: End-to-End Multi-grained Contrastive Learning for Video-Text Retrieval , by Yiwei Ma and 5 other authors", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2207.07285", "content": "Jul 15, 2022 · View a PDF of the paper titled X-CLIP: End-to-End Multi-grained Contrastive Learning for Video-Text Retrieval , by Yiwei Ma and 5 other authors"}
+{"idx": 1, "title": "X-CLIP: End-to-End Multi-grained Contrastive Learning for ...", "date": "", "ddg_snippet": "Oct 10, 2022 · Abstract Video-text retrieval has been a crucial and fundamental task in multi-modal research. The development of video-text retrieval has been considerably promoted by large-scale multi-modal contrastive pre-training, which primarily focuses on coarse-grained or fine-grained contrast.", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/abs/10.1145/3503161.3547910", "content": "Oct 10, 2022 · Abstract Video-text retrieval has been a crucial and fundamental task in multi-modal research. The development of video-text retrieval has been considerably promoted by large-scale multi-modal contrastive pre-training, which primarily focuses on coarse-grained or fine-grained contrast."}
+{"idx": 2, "title": "X-CLIP/README.md at main · xuguohai/X-CLIP · GitHub", "date": "", "ddg_snippet": "Sep 20, 2022 · The implementation of paper X-CLIP: End-to-End Multi-grained Contrastive Learning for Video-Text Retrieval . Accepted by ACMMM22. By Yiwei Ma, Guohai Xu, Xiaoshuai Sun *, Ming Yan, Ji Zhang, Rongrong Ji. X-CLIP adopts cross-grained contrastive learning and attention over similarity matrix module to filter out unnecessary information during video-text retrieval. It achieves SOTA results on MSR ...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/xuguohai/X-CLIP/blob/main/README.md", "content": "Sep 20, 2022 · The implementation of paper X-CLIP: End-to-End Multi-grained Contrastive Learning for Video-Text Retrieval . Accepted by ACMMM22. By Yiwei Ma, Guohai Xu, Xiaoshuai Sun *, Ming Yan, Ji Zhang, Rongrong Ji. X-CLIP adopts cross-grained contrastive learning and attention over similarity matrix module to filter out unnecessary information during video-text retrieval. It achieves SOTA results on MSR ..."}
+{"idx": 3, "title": "X-CLIP - Project Page - GitHub Pages", "date": "", "ddg_snippet": "Abstract Video-text retrieval has been a crucial and fundamental task in multi-modal research. The development of video-text retrieval has been considerably promoted by large-scale multi-modal contrastive pre-training, which primarily focuses on coarse-grained or finegrained contrast.", "subpage_snippet": "", "source": "xmu-xiaoma666.github.io", "link": "https://xmu-xiaoma666.github.io/Projects/MM22_XCLIP/", "content": "Abstract Video-text retrieval has been a crucial and fundamental task in multi-modal research. The development of video-text retrieval has been considerably promoted by large-scale multi-modal contrastive pre-training, which primarily focuses on coarse-grained or finegrained contrast."}
+{"idx": 4, "title": "X-CLIP: End-to-End Multi-grained Contrastive Learning for ...", "date": "", "ddg_snippet": "ABSTRACT Video-text retrieval has been a crucial and fundamental task in multi-modal research. The development of video-text retrieval has been considerably promoted by large-scale multi-modal contrastive pre-training, which primarily focuses on coarse-grained or fine-grained contrast. However, cross-grained contrast, which is the contrast between coarse-grained representations and fine ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2207.07285", "content": "ABSTRACT Video-text retrieval has been a crucial and fundamental task in multi-modal research. The development of video-text retrieval has been considerably promoted by large-scale multi-modal contrastive pre-training, which primarily focuses on coarse-grained or fine-grained contrast. However, cross-grained contrast, which is the contrast between coarse-grained representations and fine ..."}
+{"idx": 5, "title": "X-CLIP: End-to-End Multi-grained Contrastive Learning for ...", "date": "", "ddg_snippet": "Video - text retrieval has been a crucial and fundamental task in multi -modal research. The development of video - text retrieval has been considerably promoted by ...", "subpage_snippet": "", "source": "hyper.ai", "link": "https://hyper.ai/en/papers/2207.07285", "content": "Video - text retrieval has been a crucial and fundamental task in multi -modal research. The development of video - text retrieval has been considerably promoted by ..."}
+{"idx": 6, "title": "X-CLIP: End-to-End Multi-grained Contrastive Learning for ...", "date": "", "ddg_snippet": "Oct 10, 2022 · Abstract Video-text retrieval has been a crucial and fundamental task in multi-modal research. The development of video-text retrieval has been considerably promoted by large-scale multi-modal contrastive pre-training, which primarily focuses on coarse-grained or fine-grained contrast.", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/abs/10.1145/3503161.3547910?download=true", "content": "Oct 10, 2022 · Abstract Video-text retrieval has been a crucial and fundamental task in multi-modal research. The development of video-text retrieval has been considerably promoted by large-scale multi-modal contrastive pre-training, which primarily focuses on coarse-grained or fine-grained contrast."}
+{"idx": 7, "title": "‘CLIP’ tag · Gwern.net", "date": "", "ddg_snippet": "X - CLIP : Expanding Language-Image Pretrained ... X - CLIP : End - To - End Multi - Grained Contrastive Learning for Video - Text Retrieval ”, Ma et al 2022", "subpage_snippet": "", "source": "gwern.net", "link": "https://gwern.net/doc/ai/nn/transformer/clip/index", "content": "X - CLIP : Expanding Language-Image Pretrained ... X - CLIP : End - To - End Multi - Grained Contrastive Learning for Video - Text Retrieval ”, Ma et al 2022"}
+{"idx": 8, "title": "Repeating Words for Video-Language Retrieval with", "date": "", "ddg_snippet": "Video - CLIP (Xu et al., 2021 ) , UMT-B (Li et al., 2023a ) , and VINDLU (Cheng et al., 2023 ) , in text - to - video retrieval performance on MSR ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2508.14812v1", "content": "Video - CLIP (Xu et al., 2021 ) , UMT-B (Li et al., 2023a ) , and VINDLU (Cheng et al., 2023 ) , in text - to - video retrieval performance on MSR ..."}
+{"idx": 9, "title": "Ambiguity-Restrained Text-Video Representation Learning for", "date": "", "ddg_snippet": "Partially Relevant Video Retrieval (PRVR) aims to retrieve a video where a specific segment is relevant to a given text query.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2506.07471v1", "content": "Partially Relevant Video Retrieval (PRVR) aims to retrieve a video where a specific segment is relevant to a given text query."}
diff --git a/data/sampled_jsons/YuMEUNNpeb_Position_Medical_Large_Language_Model_Benchmarks_GPT-4_accuracy_Table_1.jsonl b/data/sampled_jsons/YuMEUNNpeb_Position_Medical_Large_Language_Model_Benchmarks_GPT-4_accuracy_Table_1.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..7a0edbe56e8dd7f639837516dae43477a373276c
--- /dev/null
+++ b/data/sampled_jsons/YuMEUNNpeb_Position_Medical_Large_Language_Model_Benchmarks_GPT-4_accuracy_Table_1.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Comparative benchmarking of the DeepSeek large language model on ...", "date": "", "ddg_snippet": "The open-source DeepSeek large language model showed variable performance relative to two leading models when benchmarked on four different medical tasks, with relatively strong reasoning ...", "subpage_snippet": "", "source": "www.nature.com", "link": "https://www.nature.com/articles/s41591-025-03726-3", "content": "The open-source DeepSeek large language model showed variable performance relative to two leading models when benchmarked on four different medical tasks, with relatively strong reasoning ..."}
+{"idx": 1, "title": "PDF Capabilities of GPT-4 on Medical Challenge Problems", "date": "", "ddg_snippet": "Large language models (LLMs) have demonstrated remarkable capabilities in natural language understanding and generation across various domains, including medicine. We present a compre-hensive evaluation of GPT-4 [Ope23], a state-of-the-art LLM, on medical competency examinations and benchmark datasets. GPT-4 is a general-purpose model that is not specialized for medical prob-lems through ...", "subpage_snippet": "", "source": "www.microsoft.com", "link": "https://www.microsoft.com/en-us/research/wp-content/uploads/2023/03/GPT-4_medical_benchmarks.pdf", "content": "Large language models (LLMs) have demonstrated remarkable capabilities in natural language understanding and generation across various domains, including medicine. We present a compre-hensive evaluation of GPT-4 [Ope23], a state-of-the-art LLM, on medical competency examinations and benchmark datasets. GPT-4 is a general-purpose model that is not specialized for medical prob-lems through ..."}
+{"idx": 2, "title": "Large Language Model Benchmarks in Medical Tasks", "date": "", "ddg_snippet": "Abstract—With the increasing application of large language models (LLMs) in the medical domain, evaluating these models' performance using benchmark datasets has become crucial. This paper presents a comprehensive survey of various benchmark datasets employed in medical LLM tasks. These datasets span multiple modalities including text, image, and multimodal bench-marks , focusing on ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2410.21348", "content": "Abstract—With the increasing application of large language models (LLMs) in the medical domain, evaluating these models' performance using benchmark datasets has become crucial. This paper presents a comprehensive survey of various benchmark datasets employed in medical LLM tasks. These datasets span multiple modalities including text, image, and multimodal bench-marks , focusing on ..."}
+{"idx": 3, "title": "Large Language Models in Medicine: Applications, Challenges, and Future ...", "date": "", "ddg_snippet": "In recent years, large language models (LLMs) represented by GPT-4 have developed rapidly and performed well in various natural language processing tasks, showing great potential and transformative impact. The medical field, due to its vast data ...", "subpage_snippet": "", "source": "pmc.ncbi.nlm.nih.gov", "link": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12163604/", "content": "In recent years, large language models (LLMs) represented by GPT-4 have developed rapidly and performed well in various natural language processing tasks, showing great potential and transformative impact. The medical field, due to its vast data ..."}
+{"idx": 4, "title": "Benchmarking Open-Source Large Language Models, GPT-4 and ... - NEJM AI", "date": "", "ddg_snippet": "In recent years, significant breakthroughs have been made in the field of natural language processing, particularly with the development of large language models (LLMs). LLMs have demonstrated remarkable capabilities on benchmarks related to general medical question answering, but there are fewer data about their performance in subspecialty fields and fewer studies still comparing the many ...", "subpage_snippet": "", "source": "ai.nejm.org", "link": "https://ai.nejm.org/doi/full/10.1056/AIdbp2300092", "content": "In recent years, significant breakthroughs have been made in the field of natural language processing, particularly with the development of large language models (LLMs). LLMs have demonstrated remarkable capabilities on benchmarks related to general medical question answering, but there are fewer data about their performance in subspecialty fields and fewer studies still comparing the many ..."}
+{"idx": 5, "title": "Towards evaluating and building versatile large language models for ...", "date": "", "ddg_snippet": "In this study, we present MedS-Bench, a comprehensive benchmark to evaluate large language models (LLMs) in clinical contexts, MedS-Bench, spanning 11 high-level clinical tasks.", "subpage_snippet": "", "source": "www.nature.com", "link": "https://www.nature.com/articles/s41746-024-01390-4", "content": "In this study, we present MedS-Bench, a comprehensive benchmark to evaluate large language models (LLMs) in clinical contexts, MedS-Bench, spanning 11 high-level clinical tasks."}
+{"idx": 6, "title": "PDF HealthBench: Evaluating Large Language Models Towards Improved Human Health", "date": "", "ddg_snippet": "We present HealthBench, an open-source benchmark measuring the performance and safety of large language models in healthcare. HealthBench consists of 5,000 multi-turn conversations between a model and an individual user or healthcare professional. Responses are evaluated using conversation-specific rubrics created by 262 physicians. Unlike previous multiple-choice or short-answer benchmarks ...", "subpage_snippet": "", "source": "cdn.openai.com", "link": "https://cdn.openai.com/pdf/bd7a39d5-9e9f-47b3-903c-8b847ca650c7/healthbench_paper.pdf", "content": "We present HealthBench, an open-source benchmark measuring the performance and safety of large language models in healthcare. HealthBench consists of 5,000 multi-turn conversations between a model and an individual user or healthcare professional. Responses are evaluated using conversation-specific rubrics created by 262 physicians. Unlike previous multiple-choice or short-answer benchmarks ..."}
+{"idx": 7, "title": "A systematic evaluation of large language models for biomedical natural ...", "date": "", "ddg_snippet": "This early study reported biomedical-related results, indicating that GPT-4 achieved an accuracy of approximately 80% in the US Medical Licensing Exam (Step 1 , 2, and 3), along with an example of using GPT-4 to verify claims in a medical note.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2305.16326v3", "content": "This early study reported biomedical-related results, indicating that GPT-4 achieved an accuracy of approximately 80% in the US Medical Licensing Exam (Step 1 , 2, and 3), along with an example of using GPT-4 to verify claims in a medical note."}
+{"idx": 8, "title": "PDF Evaluation of large language model performance on the Biomedical ...", "date": "", "ddg_snippet": "Methods: We evaluated and compared four general-purpose LLMs ( GPT-4 , GPT-3.5-turbo, Flan-T5-XXL, and Zephyr-7B-Beta) and a healthcare-specific LLM (MedLLaMA-13B) on a set of 13 datasets - referred to as the Biomedical Language Understanding and Reasoning Benchmark (BLURB) - covering six commonly needed medical natural language processing tasks: named entity recognition (NER); relation ...", "subpage_snippet": "", "source": "www.medrxiv.org", "link": "https://www.medrxiv.org/content/medrxiv/early/2024/05/17/2024.05.17.24307411.full.pdf", "content": "Methods: We evaluated and compared four general-purpose LLMs ( GPT-4 , GPT-3.5-turbo, Flan-T5-XXL, and Zephyr-7B-Beta) and a healthcare-specific LLM (MedLLaMA-13B) on a set of 13 datasets - referred to as the Biomedical Language Understanding and Reasoning Benchmark (BLURB) - covering six commonly needed medical natural language processing tasks: named entity recognition (NER); relation ..."}
+{"idx": 9, "title": "The Open Medical-LLM Leaderboard: Benchmarking Large Language Models in ...", "date": "", "ddg_snippet": "We're on a journey to advance and democratize artificial intelligence through open source and open science.", "subpage_snippet": "", "source": "huggingface.co", "link": "https://huggingface.co/blog/leaderboard-medicalllm", "content": "We're on a journey to advance and democratize artificial intelligence through open source and open science."}
diff --git a/data/sampled_jsons/arxiv_2405.17618_Symmetric_Cross_Entropy_loss_Equation_6_year_2024.jsonl b/data/sampled_jsons/arxiv_2405.17618_Symmetric_Cross_Entropy_loss_Equation_6_year_2024.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..e524fb4acad5f6f2a19f3f52e69ea0fda86d306f
--- /dev/null
+++ b/data/sampled_jsons/arxiv_2405.17618_Symmetric_Cross_Entropy_loss_Equation_6_year_2024.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Abstract - arXiv.org", "date": "", "ddg_snippet": "atasets. The symmetric cross entropy (SCE) [Wang et al., 2019] we mainly refer to uses a symmetric cross entropy loss , which not only considers the flow of information from the true distribution, but also includes information in the reverse d rection. SCE works better than GCE in general, especially for data with high noi", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2405.17618", "content": "atasets. The symmetric cross entropy (SCE) [Wang et al., 2019] we mainly refer to uses a symmetric cross entropy loss , which not only considers the flow of information from the true distribution, but also includes information in the reverse d rection. SCE works better than GCE in general, especially for data with high noi"}
+{"idx": 1, "title": "Symmetric Reinforcement Learning Loss for Robust Learning on Diverse ...", "date": "", "ddg_snippet": "The symmetric cross entropy (SCE) (Wang et al., 2019) we mainly refer to uses a symmetric cross entropy loss , which not only considers the flow of information from the true distribution, but also includes information in the reverse direction. SCE works better than GCE in general, especially for data with high noise rates.", "subpage_snippet": "", "source": "ar5iv.labs.arxiv.org", "link": "https://ar5iv.labs.arxiv.org/html/2405.17618", "content": "The symmetric cross entropy (SCE) (Wang et al., 2019) we mainly refer to uses a symmetric cross entropy loss , which not only considers the flow of information from the true distribution, but also includes information in the reverse direction. SCE works better than GCE in general, especially for data with high noise rates."}
+{"idx": 2, "title": "Symmetric Reinforcement Learning Loss for Robust Learning on Diverse ...", "date": "", "ddg_snippet": "In this work, we improve the stability of RL training by adapting the reverse cross entropy (RCE) from supervised learning for noisy data to define a symmetric RL loss . We demonstrate performance improvements across various tasks and scales.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2405.17618", "content": "In this work, we improve the stability of RL training by adapting the reverse cross entropy (RCE) from supervised learning for noisy data to define a symmetric RL loss . We demonstrate performance improvements across various tasks and scales."}
+{"idx": 3, "title": "(PDF) Black hole entropy in higher curvature gravity", "date": "", "ddg_snippet": "We discuss some recent results on black hole thermodynamics within the context of effective gravitational actions including higher-curvature interactions. Wald's derivation of the First Law demonstrates that black hole entropy can always be", "subpage_snippet": "", "source": "www.academia.edu", "link": "https://www.academia.edu/144058163/Black_hole_entropy_in_higher_curvature_gravity", "content": "We discuss some recent results on black hole thermodynamics within the context of effective gravitational actions including higher-curvature interactions. Wald's derivation of the First Law demonstrates that black hole entropy can always be"}
+{"idx": 4, "title": "BBANsh: a deep learning architecture based on BERT and bilinear ...", "date": "", "ddg_snippet": "The BBANsh has achieved an area under the precision-recall curve of 0.951 on five- cross validation and a prediction accuracy of 0.896 on a new external validation set, highlighting its superior ...", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/395037769_BBANsh_a_deep_learning_architecture_based_on_BERT_and_bilinear_attention_networks_to_identify_potent_shRNA", "content": "The BBANsh has achieved an area under the precision-recall curve of 0.951 on five- cross validation and a prediction accuracy of 0.896 on a new external validation set, highlighting its superior ..."}
+{"idx": 5, "title": "Enhancing TextGCN for depression detection on social media ... - Frontiers", "date": "", "ddg_snippet": "By constructing a graph of words and documents, we then pass it through a two-layer Graph Convolutional Network (GCN) designed for depression detection. Utilizing categorical cross - entropy as the loss function, we employ single-label classification to achieve accurate depression detection outcomes.", "subpage_snippet": "", "source": "www.frontiersin.org", "link": "https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2025.1612769/full", "content": "By constructing a graph of words and documents, we then pass it through a two-layer Graph Convolutional Network (GCN) designed for depression detection. Utilizing categorical cross - entropy as the loss function, we employ single-label classification to achieve accurate depression detection outcomes."}
+{"idx": 6, "title": "Performance prediction of ferrofluid-actuated double tube heat ...", "date": "", "ddg_snippet": "Enhancing heat exchanger efficiency is critical for energy conservation across industries. This study investigates the thermal-hydraulic performance and entropy generation of a double-tube heat exchanger (DTHE) utilizing magnetically actuated ferrofluid (FF). Through validated computational fluid dynamics (CFD) simulations, the impacts of magnetic source parameters (number, distance, intensity ...", "subpage_snippet": "", "source": "www.sciencedirect.com", "link": "https://www.sciencedirect.com/science/article/pii/S1359431125028200", "content": "Enhancing heat exchanger efficiency is critical for energy conservation across industries. This study investigates the thermal-hydraulic performance and entropy generation of a double-tube heat exchanger (DTHE) utilizing magnetically actuated ferrofluid (FF). Through validated computational fluid dynamics (CFD) simulations, the impacts of magnetic source parameters (number, distance, intensity ..."}
+{"idx": 7, "title": "Spatio-Temporal Recursive Method for Traffic Flow Interpolation", "date": "", "ddg_snippet": "Traffic data sequence imputation plays a crucial role in maintaining the integrity and reliability of transportation analytics and decision-making systems. With the proliferation of sensor technologies and IoT devices, traffic data often contain missing values due to sensor failures, communication issues, or data processing errors. It is necessary to effectively interpolate these missing parts ...", "subpage_snippet": "", "source": "www.mdpi.com", "link": "https://www.mdpi.com/2073-8994/17/9/1577", "content": "Traffic data sequence imputation plays a crucial role in maintaining the integrity and reliability of transportation analytics and decision-making systems. With the proliferation of sensor technologies and IoT devices, traffic data often contain missing values due to sensor failures, communication issues, or data processing errors. It is necessary to effectively interpolate these missing parts ..."}
+{"idx": 8, "title": "Self-Correcting Gottesman-Kitaev-Preskill Qubit and Gates in a Driven ...", "date": "", "ddg_snippet": "A new design for a superconducting qubit that can absorb noise-induced fluctuations may provide a path to qubit protection that avoids physical qubit overhead of quantum error-correction codes.", "subpage_snippet": "", "source": "journals.aps.org", "link": "https://journals.aps.org/prxquantum/abstract/10.1103/ykqb-m52z", "content": "A new design for a superconducting qubit that can absorb noise-induced fluctuations may provide a path to qubit protection that avoids physical qubit overhead of quantum error-correction codes."}
+{"idx": 9, "title": "MSGS-SLAM: Monocular Semantic Gaussian Splatting SLAM", "date": "", "ddg_snippet": "With the iterative evolution of SLAM (Simultaneous Localization and Mapping) technology in the robotics domain, the SLAM paradigm based on three-dimensional Gaussian distribution models has emerged as the current state-of-the-art technical approach. This research proposes a novel MSGS-SLAM system (Monocular Semantic Gaussian Splatting SLAM), which innovatively integrates monocular vision with ...", "subpage_snippet": "", "source": "www.mdpi.com", "link": "https://www.mdpi.com/2073-8994/17/9/1576", "content": "With the iterative evolution of SLAM (Simultaneous Localization and Mapping) technology in the robotics domain, the SLAM paradigm based on three-dimensional Gaussian distribution models has emerged as the current state-of-the-art technical approach. This research proposes a novel MSGS-SLAM system (Monocular Semantic Gaussian Splatting SLAM), which innovatively integrates monocular vision with ..."}
diff --git a/data/sampled_jsons/attention_head_pruning_selective_attention_language_models_efficient_inference.jsonl b/data/sampled_jsons/attention_head_pruning_selective_attention_language_models_efficient_inference.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..95a1986519be123dda4864e811e70590010fd964
--- /dev/null
+++ b/data/sampled_jsons/attention_head_pruning_selective_attention_language_models_efficient_inference.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Attention Pruning: Automated Fairness Repair of Language ...", "date": "", "ddg_snippet": "20 Mar 2025 — This paper explores pruning attention heads as a post-processing bias mitigation method for large language models (LLMs).", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2503.15815v1", "content": "20 Mar 2025 — This paper explores pruning attention heads as a post-processing bias mitigation method for large language models (LLMs)."}
+{"idx": 1, "title": "Selective Attention Improves Transformer", "date": "", "ddg_snippet": "3 Oct 2024 — Dynamic context pruning (DCP) (Anagnostidis et al., 2024) proposed a mechanism for fine tuning existing models to make inference more efficient .", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2410.02703v1", "content": "3 Oct 2024 — Dynamic context pruning (DCP) (Anagnostidis et al., 2024) proposed a mechanism for fine tuning existing models to make inference more efficient ."}
+{"idx": 2, "title": "Selective Attention Improves Transformer", "date": "", "ddg_snippet": "by Y Leviathan · Cited by 17 — Efficient Memory Management: The Selective Attention mechanism effectively prunes unneeded tokens , significantly reducing memory usage during inference without ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=v0FzmPCd1e", "content": "by Y Leviathan · Cited by 17 — Efficient Memory Management: The Selective Attention mechanism effectively prunes unneeded tokens , significantly reducing memory usage during inference without ..."}
+{"idx": 3, "title": "ALPS: Attention Localization and Pruning Strategy for ...", "date": "", "ddg_snippet": "by H Chen · 2025 — In this work, to enhance alignment efficiency, utilize weight parameters, and introduce reusabil- ity of identified attention heads , we propose. 17 pages", "subpage_snippet": "", "source": "aclanthology.org", "link": "https://aclanthology.org/2025.findings-acl.612.pdf", "content": "by H Chen · 2025 — In this work, to enhance alignment efficiency, utilize weight parameters, and introduce reusabil- ity of identified attention heads , we propose. 17 pages"}
+{"idx": 4, "title": "Efficient Inference of Large Language Models through ...", "date": "", "ddg_snippet": "by J Whitmore · 2025 — Structured pruning removes entire components, such as neurons, attention heads , or convolutional filters. Unlike unstructured pruning , ...", "subpage_snippet": "", "source": "www.preprints.org", "link": "https://www.preprints.org/frontend/manuscript/c99e34e92dafd86f14e148c6d8501339/download_pub", "content": "by J Whitmore · 2025 — Structured pruning removes entire components, such as neurons, attention heads , or convolutional filters. Unlike unstructured pruning , ..."}
+{"idx": 5, "title": "A Survey (Awesome Attention Heads)", "date": "", "ddg_snippet": "In this survey, we delve into the potential mechanisms of how attention heads in LLMs contribute to the reasoning process.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/IAAR-Shanghai/Awesome-Attention-Heads", "content": "In this survey, we delve into the potential mechanisms of how attention heads in LLMs contribute to the reasoning process."}
+{"idx": 6, "title": "Pruning Attention Heads with Almost-sure Sparsity Targets", "date": "", "ddg_snippet": "To push the envelope on inference efficiency, we propose a novel technique, concentrator, based on which we develop PASSCONC (PASS with CONCentrator), as ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/notes/edits/attachment?id=LGnlHdmTDg&name=pdf", "content": "To push the envelope on inference efficiency, we propose a novel technique, concentrator, based on which we develop PASSCONC (PASS with CONCentrator), as ..."}
+{"idx": 7, "title": "Sparsity in Deep Learning: Pruning and growth for efficient ...", "date": "", "ddg_snippet": "by T Hoefler · 2021 · Cited by 1106 — In this paper, we survey prior work on sparsity in deep learning and provide an extensive tutorial of sparsification for both inference and training. We ...", "subpage_snippet": "", "source": "jmlr.org", "link": "https://jmlr.org/papers/volume22/21-0366/21-0366.pdf", "content": "by T Hoefler · 2021 · Cited by 1106 — In this paper, we survey prior work on sparsity in deep learning and provide an extensive tutorial of sparsification for both inference and training. We ..."}
+{"idx": 8, "title": "Selective Attention Improves Transformer", "date": "", "ddg_snippet": "3 Oct 2024 — Selective attention enhances transformer performance by focusing on relevant elements, improving language modeling across various sizes and ...", "subpage_snippet": "", "source": "powerdrill.ai", "link": "https://powerdrill.ai/discover/discover-Selective-Attention-Improves-cm1v7mdwguur3013w3dkko94l", "content": "3 Oct 2024 — Selective attention enhances transformer performance by focusing on relevant elements, improving language modeling across various sizes and ..."}
+{"idx": 9, "title": "Selective Attention in Transformers - Navdeet Saini", "date": "", "ddg_snippet": "Efficient Pruning During Inference : Context Pruning is enabled, allowing models to remove tokens with high soft-mask values, thus reducing ...", "subpage_snippet": "", "source": "www.linkedin.com", "link": "https://www.linkedin.com/pulse/selective-attention-transformers-maximizing-efficiency-navdeet-saini-bpiqc", "content": "Efficient Pruning During Inference : Context Pruning is enabled, allowing models to remove tokens with high soft-mask values, thus reducing ..."}
diff --git a/data/sampled_jsons/extragradient_method_procedural_difference_standard_gradient_descent_two_step_process.jsonl b/data/sampled_jsons/extragradient_method_procedural_difference_standard_gradient_descent_two_step_process.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..d38838908dde8653b1daf04e2e7e7751b9112bc3
--- /dev/null
+++ b/data/sampled_jsons/extragradient_method_procedural_difference_standard_gradient_descent_two_step_process.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Extragradient Method in Optimization: Convergence and Complexity", "date": "", "ddg_snippet": "We combined the work of [5,8] and some recent extensions for rst-order descent methods , (see [9{12]), to propose the extented extragradient method (EEG for short) to tackle the problem of minimizing a com-posite objective function. The classical extragradient method relies on orthogonal projections. We extend it by considering more general nonsmooth functions, and using proximal gradient steps ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/1609.08177", "content": "We combined the work of [5,8] and some recent extensions for rst-order descent methods , (see [9{12]), to propose the extented extragradient method (EEG for short) to tackle the problem of minimizing a com-posite objective function. The classical extragradient method relies on orthogonal projections. We extend it by considering more general nonsmooth functions, and using proximal gradient steps ..."}
+{"idx": 1, "title": "Extragradient Method in Optimization: Convergence and Complexity", "date": "", "ddg_snippet": "A distinguishing feature of the extragradient method is its use of an additional projected gradient step , which can be seen as a guide during the optimization process .", "subpage_snippet": "", "source": "link.springer.com", "link": "https://link.springer.com/article/10.1007/s10957-017-1200-6", "content": "A distinguishing feature of the extragradient method is its use of an additional projected gradient step , which can be seen as a guide during the optimization process ."}
+{"idx": 2, "title": "The Extragradient Method: A Key to Complex Problem Solving", "date": "", "ddg_snippet": "Title: Revisiting Extragradient -Type Methods -- Part 1: Generalizations and Sublinear Convergence Rates Abstract: This paper presents a comprehensive analysis of the well-known extragradient (EG) method for solving both equations and inclusions. First, we unify and generalize EG for [non]linear equations to a wider class of algorithms, encompassing various existing schemes and potentially new ...", "subpage_snippet": "", "source": "scisimple.com", "link": "https://scisimple.com/en/articles/2025-05-30-the-extragradient-method-a-key-to-complex-problem-solving--a9rdnrq", "content": "Title: Revisiting Extragradient -Type Methods -- Part 1: Generalizations and Sublinear Convergence Rates Abstract: This paper presents a comprehensive analysis of the well-known extragradient (EG) method for solving both equations and inclusions. First, we unify and generalize EG for [non]linear equations to a wider class of algorithms, encompassing various existing schemes and potentially new ..."}
+{"idx": 3, "title": "What is an extragradient method? - MathOverflow", "date": "", "ddg_snippet": "The extragradient method uses another extra gradient step for the new variable and reads as $$ u^ {n+1} = P_C (u^n - \\sigma\\nabla F (\\bar u^n)),\\qquad \\bar u^ {n+1} = P_C (u^ {n+1} - \\sigma \\nabla F (\\bar u^n)) $$ The benefit of this approach is that, although you make a new gradient step and usually have \"better convergence\", you have to ...", "subpage_snippet": "", "source": "mathoverflow.net", "link": "https://mathoverflow.net/questions/179325/what-is-an-extragradient-method", "content": "The extragradient method uses another extra gradient step for the new variable and reads as $$ u^ {n+1} = P_C (u^n - \\sigma\\nabla F (\\bar u^n)),\\qquad \\bar u^ {n+1} = P_C (u^ {n+1} - \\sigma \\nabla F (\\bar u^n)) $$ The benefit of this approach is that, although you make a new gradient step and usually have \"better convergence\", you have to ..."}
+{"idx": 4, "title": "PDF Lecture 15: Extragradient - GitHub Pages", "date": "", "ddg_snippet": "\"A unified analysis of extra-gradient and optimistic gradient methods for saddle point problems: Proximal point approach.\" International Conference on Artificial Intelligence and Statistics.", "subpage_snippet": "", "source": "gauthiergidel.github.io", "link": "https://gauthiergidel.github.io/courses/slides/Lecture15.pdf", "content": "\"A unified analysis of extra-gradient and optimistic gradient methods for saddle point problems: Proximal point approach.\" International Conference on Artificial Intelligence and Statistics."}
+{"idx": 5, "title": "PDF Sublinear Convergence Rates of Extragradient-Type Methods: Classical ...", "date": "", "ddg_snippet": "This method performs two sequential gradient /forward steps at each iteration, making it twice as expensive as the standard gradient method , but convergent under only the monotonicity and the ...", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/profile/Quoc-Tran-Dinh/publication/369625810_Sublinear_Convergence_Rates_of_Extragradient-Type_Methods_A_Survey_on_Classical_and_Recent_Developments/links/646bb04127938813482df6fc/Sublinear-Convergence-Rates-of-Extragradient-Type-Methods-A-Survey-on-Classical-and-Recent-Developments.pdf", "content": "This method performs two sequential gradient /forward steps at each iteration, making it twice as expensive as the standard gradient method , but convergent under only the monotonicity and the ..."}
+{"idx": 6, "title": "Strongly convergent two-step inertial subgradient extragradient methods ...", "date": "", "ddg_snippet": "In this paper, we introduce two modified subgradient extragradient methods with two -inertial steps based on Halpern-type and Mann-type iterations for approximating solutions of variational inequalities involving quasi-monotone operators in real Hilbert spaces.", "subpage_snippet": "", "source": "www.sciencedirect.com", "link": "https://www.sciencedirect.com/science/article/pii/S1007570425003703", "content": "In this paper, we introduce two modified subgradient extragradient methods with two -inertial steps based on Halpern-type and Mann-type iterations for approximating solutions of variational inequalities involving quasi-monotone operators in real Hilbert spaces."}
+{"idx": 7, "title": "PDF Explore Aggressively, Update Conservatively: Stochastic Extragradient ...", "date": "", "ddg_snippet": "Owing to their stability and convergence speed, extragradient methods have be-come a staple for solving large-scale saddle-point problems in machine learning. The basic premise of these algorithms is the use of an extrapolation step before per-forming an update; thanks to this exploration step , extragradient methods overcome many of the non-convergence issues that plague gradient descent ...", "subpage_snippet": "", "source": "proceedings.neurips.cc", "link": "https://proceedings.neurips.cc/paper/2020/file/ba9a56ce0a9bfa26e8ed9e10b2cc8f46-Paper.pdf", "content": "Owing to their stability and convergence speed, extragradient methods have be-come a staple for solving large-scale saddle-point problems in machine learning. The basic premise of these algorithms is the use of an extrapolation step before per-forming an update; thanks to this exploration step , extragradient methods overcome many of the non-convergence issues that plague gradient descent ..."}
+{"idx": 8, "title": "Double-Step Alternating Extragradient with Increasing Timescale ...", "date": "", "ddg_snippet": "Abstract In nonconvex-nonconcave minimax optimization, two -timescale gradient methods have shown their potential to find local minimax (optimal) points, provided that the timescale separation between the min and the max player is suficiently large. However, existing two -timescale variants of gradi-ent descent ascent and extragradient methods face two shortcomings, especially when we search for ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=nUVForc3VP", "content": "Abstract In nonconvex-nonconcave minimax optimization, two -timescale gradient methods have shown their potential to find local minimax (optimal) points, provided that the timescale separation between the min and the max player is suficiently large. However, existing two -timescale variants of gradi-ent descent ascent and extragradient methods face two shortcomings, especially when we search for ..."}
+{"idx": 9, "title": "Subgradient Extragradient Method with Double Inertial Steps for ...", "date": "", "ddg_snippet": "Our proposed method is a combination of double inertial extrapolation steps , relaxation step and subgradient extragradient method which is aimed to increase the speed of convergence of many available subgradient extragradient methods with inertia for solving variational inequalities.", "subpage_snippet": "", "source": "link.springer.com", "link": "https://link.springer.com/article/10.1007/s10915-021-01751-1", "content": "Our proposed method is a combination of double inertial extrapolation steps , relaxation step and subgradient extragradient method which is aimed to increase the speed of convergence of many available subgradient extragradient methods with inertia for solving variational inequalities."}
diff --git a/data/sampled_jsons/httpswww.researchgate.netpublication388658326_Blink_of_an_eye_a_simple_theory_for_feature_localizati.jsonl b/data/sampled_jsons/httpswww.researchgate.netpublication388658326_Blink_of_an_eye_a_simple_theory_for_feature_localizati.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..aa69897740e8e2703ae7b0dd6b6a2bd9bb9735f4
--- /dev/null
+++ b/data/sampled_jsons/httpswww.researchgate.netpublication388658326_Blink_of_an_eye_a_simple_theory_for_feature_localizati.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "(PDF) Blink of an eye: a simple theory for feature ...", "date": "", "ddg_snippet": "Feb 2, 2025 · Large language models (LLMs) can exhibit undesirable and unexpected behavior in the blink of an eye . In a recent Anthropic demo, Claude switched from coding to Googling pictures of Yellowstone ...", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/388658326_Blink_of_an_eye_a_simple_theory_for_feature_localization_in_generative_models", "content": "Feb 2, 2025 · Large language models (LLMs) can exhibit undesirable and unexpected behavior in the blink of an eye . In a recent Anthropic demo, Claude switched from coding to Googling pictures of Yellowstone ..."}
+{"idx": 1, "title": "Blink of an eye: a simple theory for feature localization in ...", "date": "", "ddg_snippet": "Feb 2, 2025 · Large language models can exhibit unexpected behavior in the blink of an eye . In a recent computer use demo, a language model switched from coding to Googling pictures of Yellowstone, and these sudden shifts in behavior have also been observed in reasoning patterns and jailbreaks. This phenomenon is not unique to autoregressive models : in diffusion models , key features of the final output are ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2502.00921", "content": "Feb 2, 2025 · Large language models can exhibit unexpected behavior in the blink of an eye . In a recent computer use demo, a language model switched from coding to Googling pictures of Yellowstone, and these sudden shifts in behavior have also been observed in reasoning patterns and jailbreaks. This phenomenon is not unique to autoregressive models : in diffusion models , key features of the final output are ..."}
+{"idx": 2, "title": "dblp: Blink of an eye: a simple theory for feature ...", "date": "", "ddg_snippet": "Mar 7, 2025 · Bibliographic details on Blink of an eye : a simple theory for feature localization in generative models .", "subpage_snippet": "", "source": "dblp.org", "link": "https://dblp.org/rec/journals/corr/abs-2502-00921", "content": "Mar 7, 2025 · Bibliographic details on Blink of an eye : a simple theory for feature localization in generative models ."}
+{"idx": 3, "title": "Blink of an eye: a simple theory for feature localization in...", "date": "", "ddg_snippet": "May 1, 2025 · In this work we develop a simple , unifying theory to explain this phenomenon. Using the formalism of stochastic localization for generative models , we show that it emerges generically as the generation process localizes to a sub-population of the distribution it models .", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=QvqnPVGWAN", "content": "May 1, 2025 · In this work we develop a simple , unifying theory to explain this phenomenon. Using the formalism of stochastic localization for generative models , we show that it emerges generically as the generation process localizes to a sub-population of the distribution it models ."}
+{"idx": 4, "title": "ResearchGate", "date": "", "ddg_snippet": "ResearchGate", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/388658326_Blink_of_an_eye_a_simple_theory_for_feature_localization_in_generative_models/fulltext/67a191718311ce680c5059c3/Blink-of-an-eye-a-simple-theory-for-feature-localization-in-generative-models.pdf", "content": "ResearchGate"}
+{"idx": 5, "title": "Blink of an eye: a simple theory for feature localization in ...", "date": "", "ddg_snippet": "To study feature localization in diffusion and autoregressive models , we consider a forward-reverseexperiment. A forward-reverse experiment considers the amount of “noise” one would need to add to a generation so that running the generative model starting from the noised generation would still yield a sample with the same feature .", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2502.00921v1", "content": "To study feature localization in diffusion and autoregressive models , we consider a forward-reverseexperiment. A forward-reverse experiment considers the amount of “noise” one would need to add to a generation so that running the generative model starting from the noised generation would still yield a sample with the same feature ."}
+{"idx": 6, "title": "Feature Information Driven Position Gaussian Distribution Estimation...", "date": "", "ddg_snippet": "...Bian and others published Feature Information Driven Position Gaussian Distribution Estimation for Tiny Object Detection | Find, read and cite all the research you need on ResearchGate .", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/394633809_Feature_Information_Driven_Position_Gaussian_Distribution_Estimation_for_Tiny_Object_Detection", "content": "...Bian and others published Feature Information Driven Position Gaussian Distribution Estimation for Tiny Object Detection | Find, read and cite all the research you need on ResearchGate ."}
+{"idx": 7, "title": "在线 PDF 添加水印 - 在线超级转换工具", "date": "", "ddg_snippet": "PDF 转换...", "subpage_snippet": "", "source": "pdf.wdku.net", "link": "https://pdf.wdku.net/pdf_add_watermark", "content": "PDF 转换..."}
+{"idx": 8, "title": "Вход в ваш аккаунт | Hotgens. net", "date": "", "ddg_snippet": "Войдите на Hotgens. net , чтобы получить доступ к мощному NSFW AI-генератору изображений.", "subpage_snippet": "", "source": "hotgens.net", "link": "https://hotgens.net/ru/auth/signin", "content": "Войдите на Hotgens. net , чтобы получить доступ к мощному NSFW AI-генератору изображений."}
+{"idx": 9, "title": "Homepage: Sitan Chen", "date": "", "ddg_snippet": "Blink of an Eye : A Simple Theory for Feature Localization in Generative Models [ pdf ] Marvin Li, Aayush Karan, Sitan Chen ICML 2025 Oral presentation Gradient Dynamics for Low-Rank Fine-Tuning Beyond Kernels [ pdf ] [slides] Arif Kerem Dayi, Sitan Chen COLT 2025 Predicting Quantum Channels Over General Product Distributions [ pdf ]", "subpage_snippet": "", "source": "www.sitanchen.com", "link": "https://www.sitanchen.com/", "content": "Blink of an Eye : A Simple Theory for Feature Localization in Generative Models [ pdf ] Marvin Li, Aayush Karan, Sitan Chen ICML 2025 Oral presentation Gradient Dynamics for Low-Rank Fine-Tuning Beyond Kernels [ pdf ] [slides] Arif Kerem Dayi, Sitan Chen COLT 2025 Predicting Quantum Channels Over General Product Distributions [ pdf ]"}
diff --git a/data/sampled_jsons/learnable_PDE_solvers_multi-time-step_schemes_error_correction_comparison.jsonl b/data/sampled_jsons/learnable_PDE_solvers_multi-time-step_schemes_error_correction_comparison.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..f2fc9416b216ec7ce8b160309f66ee94c41292d9
--- /dev/null
+++ b/data/sampled_jsons/learnable_PDE_solvers_multi-time-step_schemes_error_correction_comparison.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "(PDF) A Neural PDE Solver with Temporal Stencil Modeling", "date": "", "ddg_snippet": "The power of learnable PDE solvers is usually believ ed to.in each method, the divergence operator , the explicit time - step operator, and the pressure projection (in yellow color)are. shared between classic solvers and learnable methods.", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/368572925_A_Neural_PDE_Solver_with_Temporal_Stencil_Modeling", "content": "The power of learnable PDE solvers is usually believ ed to.in each method, the divergence operator , the explicit time - step operator, and the pressure projection (in yellow color)are. shared between classic solvers and learnable methods."}
+{"idx": 1, "title": "When solving partial differential equations ( PDEs ), classical...", "date": "", "ddg_snippet": "Using the Euler scheme for time stepping instead of RK4 results in reduced stability. and increased error accumulation, leading to a decrease in performance compared to the complete. P2C2Net. Overall, the results confirm that the physics-encoded variable correction learning method.", "subpage_snippet": "", "source": "papers.nips.cc", "link": "https://papers.nips.cc/paper_files/paper/2024/file/7f605d59a0dbde101518b552cb616ddf-Paper-Conference.pdf", "content": "Using the Euler scheme for time stepping instead of RK4 results in reduced stability. and increased error accumulation, leading to a decrease in performance compared to the complete. P2C2Net. Overall, the results confirm that the physics-encoded variable correction learning method."}
+{"idx": 2, "title": "Deep backward multistep schemes for nonlinear PDEs and...", "date": "", "ddg_snippet": "We develop multistep machine learning schemes for solving nonlinear partial differential equations ( PDEs ) in high dimension. The method is based on probabilistic representation of PDEs by backward stochastic differential equations (BSDEs) and its iterated time discretization.", "subpage_snippet": "", "source": "fime-lab.org", "link": "https://fime-lab.org/deep-backward-multistep-schemes-for-nonlinear-pdes-and-approximation-error-analysis-m-germain-h-pham-x-warin/", "content": "We develop multistep machine learning schemes for solving nonlinear partial differential equations ( PDEs ) in high dimension. The method is based on probabilistic representation of PDEs by backward stochastic differential equations (BSDEs) and its iterated time discretization."}
+{"idx": 3, "title": "Toward General-Purpose Monte Carlo PDE Solvers for Graphics...", "date": "", "ddg_snippet": "This thesis develops novel Monte Carlo methods for solving a wide range of partial differential equations ( PDEs ) relevant to computer graphics.", "subpage_snippet": "", "source": "uwspace.uwaterloo.ca", "link": "https://uwspace.uwaterloo.ca/items/65467eb5-a193-45ed-8e53-ffe267c411c7", "content": "This thesis develops novel Monte Carlo methods for solving a wide range of partial differential equations ( PDEs ) relevant to computer graphics."}
+{"idx": 4, "title": "A multi time - step partitioned approach for the coupling of SPH and FE...", "date": "", "ddg_snippet": "point scheme will be employed. As this integration scheme requires the determination of the uid status at an intermediate stage (tn+1/2) to predict the corrected solution at the nal stage (tn+1), the system (2.24) must be expressed at both time stages", "subpage_snippet": "", "source": "theses.hal.science", "link": "https://theses.hal.science/tel-01977755/document", "content": "point scheme will be employed. As this integration scheme requires the determination of the uid status at an intermediate stage (tn+1/2) to predict the corrected solution at the nal stage (tn+1), the system (2.24) must be expressed at both time stages"}
+{"idx": 5, "title": "Multi - time - step domain decomposition method with non-matching...", "date": "", "ddg_snippet": "Parabolic problems are usually solved by discretizing spatially using finite elements and then integrating over time using discrete solvers . We propose an asynchronous multi -domain time integration scheme for parabolic problems.", "subpage_snippet": "", "source": "ideas.repec.org", "link": "https://ideas.repec.org/a/eee/apmaco/v267y2015icp571-582.html", "content": "Parabolic problems are usually solved by discretizing spatially using finite elements and then integrating over time using discrete solvers . We propose an asynchronous multi -domain time integration scheme for parabolic problems."}
+{"idx": 6, "title": "Asynchronous multi -domain variational integrators for nonlinear...", "date": "", "ddg_snippet": "Typically, the partial differential equations ( PDEs ) are solved by discretizing the domain spatially using nite elements or nite volumes, then integrating over time using a numerical ordinary differential equation (ODE) solver .", "subpage_snippet": "", "source": "www3.nd.edu", "link": "https://www3.nd.edu/~kmatous/Papers/CMAME_AMVI_PDE.pdf", "content": "Typically, the partial differential equations ( PDEs ) are solved by discretizing the domain spatially using nite elements or nite volumes, then integrating over time using a numerical ordinary differential equation (ODE) solver ."}
+{"idx": 7, "title": "Alias-Free Mamba Neural Operator", "date": "", "ddg_snippet": "Learnable PDE Solvers . Alias-free Framework. Mamba Neural Operator.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=gUEBXGV8JM", "content": "Learnable PDE Solvers . Alias-free Framework. Mamba Neural Operator."}
+{"idx": 8, "title": "DimINO: Dimension-Informed Neural Operator Learning", "date": "", "ddg_snippet": "Learning-Based PDE Solvers . error on the test set compared to the baseline. This result further demonstrates the effectiveness of DimINO’s adaptive capabilities in capturing the dynamics of nonlinear PDEs .", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2410.05894v4", "content": "Learning-Based PDE Solvers . error on the test set compared to the baseline. This result further demonstrates the effectiveness of DimINO’s adaptive capabilities in capturing the dynamics of nonlinear PDEs ."}
+{"idx": 9, "title": "Daily Papers - Hugging Face", "date": "", "ddg_snippet": "...parameter-efficient solver (e.g., with 80 learnable parameters), is trained for roughly 1% of the GPU time required for training the pre-trained model, and significantly improves approximation and generation quality compared to dedicated solvers .", "subpage_snippet": "", "source": "huggingface.co", "link": "https://huggingface.co/papers?q=DiffusionPDE+solvers", "content": "...parameter-efficient solver (e.g., with 80 learnable parameters), is trained for roughly 1% of the GPU time required for training the pre-trained model, and significantly improves approximation and generation quality compared to dedicated solvers ."}
diff --git a/data/sampled_jsons/machine_unlearning_survey_per-instance_privacy.jsonl b/data/sampled_jsons/machine_unlearning_survey_per-instance_privacy.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..e609608b8469b73df06372b8b301d424f2b94b15
--- /dev/null
+++ b/data/sampled_jsons/machine_unlearning_survey_per-instance_privacy.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Leveraging Per - Instance Privacy for Machine Unlearning", "date": "", "ddg_snippet": "• Per - Instance Theoretical Analysis of Unlearning .(2021) proved that, under various assumptions, their. 2. Leveraging Per - Instance Privacy for Machine Unlearning . approach achieves approximate unlearning in a sequential framework with fixed per-deletion runtime.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2505.18786", "content": "• Per - Instance Theoretical Analysis of Unlearning .(2021) proved that, under various assumptions, their. 2. Leveraging Per - Instance Privacy for Machine Unlearning . approach achieves approximate unlearning in a sequential framework with fixed per-deletion runtime."}
+{"idx": 1, "title": "Leveraging Per -Example Privacy for Machine Unlearning", "date": "", "ddg_snippet": "Unlike other unlearning methods that lack theoretical guarantees for non-convex models, our approach builds on recent advances in differential privacy to provide per - instance guarantees using Rényi divergence.", "subpage_snippet": "", "source": "research.google", "link": "https://research.google/pubs/leveraging-per-example-privacy-for-machine-unlearning/", "content": "Unlike other unlearning methods that lack theoretical guarantees for non-convex models, our approach builds on recent advances in differential privacy to provide per - instance guarantees using Rényi divergence."}
+{"idx": 2, "title": "Frontiers in Machine Learning: Synthesizing May... - DEV Community", "date": "", "ddg_snippet": "Per - Instance Machine Unlearning Enables Targeted Data Removal Advances in machine unlearning provide tools for efficient and fair removal of specific data influences from trained models. By quantifying per - instance privacy loss and enabling adaptive updates...", "subpage_snippet": "", "source": "dev.to", "link": "https://dev.to/khanali21/frontiers-in-machine-learning-synthesizing-may-2025-arxiv-cslg-advances-in-efficiency-18o5", "content": "Per - Instance Machine Unlearning Enables Targeted Data Removal Advances in machine unlearning provide tools for efficient and fair removal of specific data influences from trained models. By quantifying per - instance privacy loss and enabling adaptive updates..."}
+{"idx": 3, "title": "GitHub - tamlhp/awesome- machine - unlearning : Awesome Machine ...", "date": "", "ddg_snippet": "2025. A Survey of Machine Unlearning . instance -wise unlearning . [Code]. Parameter-tuning-free data entry error unlearning with adaptive selective synaptic dampening.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/tamlhp/awesome-machine-unlearning", "content": "2025. A Survey of Machine Unlearning . instance -wise unlearning . [Code]. Parameter-tuning-free data entry error unlearning with adaptive selective synaptic dampening."}
+{"idx": 4, "title": "CS PhD, UofT - Cited by 775 - Machine Learning - Computer Security", "date": "", "ddg_snippet": "Leveraging Per - Instance Privacy for Machine Unlearning . NM Sepahvand, A Thudi, B Isik, A Bhattacharyya, N PapernotICML 2025 Workshop on Machine Unlearning for Generative AI, 0. The system can't perform the operation now. Try again later.", "subpage_snippet": "", "source": "scholar.google.com", "link": "https://scholar.google.com/citations?user=bTEybH0AAAAJ&hl=en", "content": "Leveraging Per - Instance Privacy for Machine Unlearning . NM Sepahvand, A Thudi, B Isik, A Bhattacharyya, N PapernotICML 2025 Workshop on Machine Unlearning for Generative AI, 0. The system can't perform the operation now. Try again later."}
+{"idx": 5, "title": "Machine Unlearning : A key to privacy in AI, but at what cost?", "date": "", "ddg_snippet": "Privacy vulnerabilities in unlearning pipelines. Beyond security, machine unlearning raises significant privacy concerns.", "subpage_snippet": "", "source": "www.devdiscourse.com", "link": "https://www.devdiscourse.com/article/technology/3224182-machine-unlearning-a-key-to-privacy-in-ai-but-at-what-cost", "content": "Privacy vulnerabilities in unlearning pipelines. Beyond security, machine unlearning raises significant privacy concerns."}
+{"idx": 6, "title": "(PDF) A Survey of Machine Unlearning", "date": "", "ddg_snippet": "The survey highlights that significant research into machine unlearning started around 2019, as datasets and models began expanding considerably.", "subpage_snippet": "", "source": "www.academia.edu", "link": "https://www.academia.edu/118741797/A_Survey_of_Machine_Unlearning", "content": "The survey highlights that significant research into machine unlearning started around 2019, as datasets and models began expanding considerably."}
+{"idx": 7, "title": "Machine Unlearning : A Survey | ACM Computing Surveys", "date": "", "ddg_snippet": "Machine Unlearning : A Survey . Authors: Heng Xu, Tianqing Zhu, Lefeng Zhang, Wanlei Zhou, Philip S. YuAuthors Info & Claims.2021. When machine unlearning jeopardizes privacy . In ACM SIGSAC Conference on Computer and Communications Security.", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/full/10.1145/3603620?cookieSet=1", "content": "Machine Unlearning : A Survey . Authors: Heng Xu, Tianqing Zhu, Lefeng Zhang, Wanlei Zhou, Philip S. YuAuthors Info & Claims.2021. When machine unlearning jeopardizes privacy . In ACM SIGSAC Conference on Computer and Communications Security."}
+{"idx": 8, "title": "(PDF) Comparative Study of Machine Unlearning Techniques for...", "date": "", "ddg_snippet": "Machine unlearning survey . Conference Paper. Machine unlearning , an emerging research topic focusing on compliance with data privacy regulations, enables trained models to remove the information learned from specific data.", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/387930995_Comparative_Study_of_Machine_Unlearning_Techniques_for_Computer_Vision_and_NLP_Models", "content": "Machine unlearning survey . Conference Paper. Machine unlearning , an emerging research topic focusing on compliance with data privacy regulations, enables trained models to remove the information learned from specific data."}
+{"idx": 9, "title": "Learn to Unlearn : A Survey on Machine Unlearning -Bohrium", "date": "", "ddg_snippet": "Overall, the survey provides a thorough synopsis of machine unlearning techniques and applications, noting future research directions in this evolving field. The survey aims to be a valuable resource for researchers and practitioners seeking to provide privacy and equity in ML systems.", "subpage_snippet": "", "source": "www.bohrium.com", "link": "https://www.bohrium.com/paper-details/learn-to-unlearn-a-survey-on-machine-unlearning/867791164751741900-108614", "content": "Overall, the survey provides a thorough synopsis of machine unlearning techniques and applications, noting future research directions in this evolving field. The survey aims to be a valuable resource for researchers and practitioners seeking to provide privacy and equity in ML systems."}
diff --git a/data/sampled_jsons/medium_learning_rate_small_learning_rate_5e-5_1e-5_fine-tuning_year_2024.jsonl b/data/sampled_jsons/medium_learning_rate_small_learning_rate_5e-5_1e-5_fine-tuning_year_2024.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..005073c494825c43171dee38cbb4af2b4c4dfaae
--- /dev/null
+++ b/data/sampled_jsons/medium_learning_rate_small_learning_rate_5e-5_1e-5_fine-tuning_year_2024.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Stable Gradients for Stable Learning at Scale in Deep", "date": "", "ddg_snippet": "A deep reinforcement learning agent interacts with an environment through sequences of actions ( a ∈ 𝒜 𝑎 𝒜 a\\in\\mathcal{A} italic_a ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2506.15544v1", "content": "A deep reinforcement learning agent interacts with an environment through sequences of actions ( a ∈ 𝒜 𝑎 𝒜 a\\in\\mathcal{A} italic_a ..."}
+{"idx": 1, "title": "LLM Fine-Tuning: Complete Guide to Optimizing Language Models", "date": "", "ddg_snippet": "Key parameters include learning rate (typically 1e - 5 to 5e - 5 ), batch size, and epochs. ... Your learning rate should be lower than the original pre ...", "subpage_snippet": "", "source": "www.spaceo.ai", "link": "https://www.spaceo.ai/blog/llm-fine-tuning/", "content": "Key parameters include learning rate (typically 1e - 5 to 5e - 5 ), batch size, and epochs. ... Your learning rate should be lower than the original pre ..."}
+{"idx": 2, "title": "A comparison of pipelines for the translation of a low resource", "date": "", "ddg_snippet": "Through the fine-tuning of large-language models on multilingual datasets, the use of specific corpora, and transfer learning techniques, the project ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2509.12514v1", "content": "Through the fine-tuning of large-language models on multilingual datasets, the use of specific corpora, and transfer learning techniques, the project ..."}
+{"idx": 3, "title": "OWSM v3.1: Better and Faster Open Whisper-Style Speech Models", "date": "", "ddg_snippet": "Our base, small , and medium models ... A typical strategy to improve convergence is to use a very small learning rate at the beginning of training.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2401.16658v3", "content": "Our base, small , and medium models ... A typical strategy to improve convergence is to use a very small learning rate at the beginning of training."}
+{"idx": 4, "title": "Flexible Tool Selection through Low-dimensional Attribute", "date": "", "ddg_snippet": "... in tool selection tasks, substantially outperforming direct tool name matching (20%) and smaller multimodal large language models (LLMs) (21%-58 ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2505.22146v4", "content": "... in tool selection tasks, substantially outperforming direct tool name matching (20%) and smaller multimodal large language models (LLMs) (21%-58 ..."}
+{"idx": 5, "title": "LoRA training guide Version 3! I go more in-depth with datasets", "date": "", "ddg_snippet": "... a question about learning rate , if you increase batch-size, should the learning rate ... So i would go for unet at 1e -4 and learning rate at 5e - 5 .", "subpage_snippet": "", "source": "www.reddit.com", "link": "https://www.reddit.com/r/StableDiffusion/comments/11vw5k3/lora_training_guide_version_3_i_go_more_indepth/", "content": "... a question about learning rate , if you increase batch-size, should the learning rate ... So i would go for unet at 1e -4 and learning rate at 5e - 5 ."}
+{"idx": 6, "title": "GitHub - NExTplusplus/TAT-QA: TAT-QA (Tabular And Textual", "date": "", "ddg_snippet": "... 50 --warmup 0.06 --optimizer adam -- learning _ rate 5e -4 \\ --weight_decay 5e - 5 --seed 123 --gradient_accumulation_steps 4 --bert_ learning _ rate 1 . 5e - 5 ...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/NExTplusplus/tat-qa", "content": "... 50 --warmup 0.06 --optimizer adam -- learning _ rate 5e -4 \\ --weight_decay 5e - 5 --seed 123 --gradient_accumulation_steps 4 --bert_ learning _ rate 1 . 5e - 5 ..."}
+{"idx": 7, "title": "Quickstart with Python", "date": "", "ddg_snippet": "... sft \" , epochs= 3 , batch_size= 1 , lr= 1e - 5 ... We are training the model for 3 epochs with a batch size of 1 and a learning rate of 1e - 5 .", "subpage_snippet": "", "source": "huggingface.co", "link": "https://huggingface.co/docs/autotrain/quickstart_py", "content": "... sft \" , epochs= 3 , batch_size= 1 , lr= 1e - 5 ... We are training the model for 3 epochs with a batch size of 1 and a learning rate of 1e - 5 ."}
+{"idx": 8, "title": "Continuing Pre-Training on Raw Text · Chris McCormick", "date": "", "ddg_snippet": "So domain knowledge includes things like learning about new “entities” (people, organizations, projects, …), and new “jargon”.", "subpage_snippet": "", "source": "mccormickml.com", "link": "http://mccormickml.com/2025/01/18/continuing-pre-training-on-raw-text/", "content": "So domain knowledge includes things like learning about new “entities” (people, organizations, projects, …), and new “jargon”."}
+{"idx": 9, "title": "", "date": "", "ddg_snippet": "", "subpage_snippet": "", "source": "", "link": "", "content": ""}
diff --git a/data/sampled_jsons/neural_PDE_solver_multiscale_time_stepping_error_correction.jsonl b/data/sampled_jsons/neural_PDE_solver_multiscale_time_stepping_error_correction.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..9be428a6d47fb66ca625e6d786d32d8c1be4c13d
--- /dev/null
+++ b/data/sampled_jsons/neural_PDE_solver_multiscale_time_stepping_error_correction.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "MultiPDENet: PDE-embedded Learning with Multi-time-stepping", "date": "", "ddg_snippet": "Notably, it integrates a trainable neural solver for precise predictions at micro time scales, while employing a NN to correct errors at macro time steps. Additionally, by embedding PDEs , MultiPDENet offers enhanced generalizability.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2501.15987v1", "content": "Notably, it integrates a trainable neural solver for precise predictions at micro time scales, while employing a NN to correct errors at macro time steps. Additionally, by embedding PDEs , MultiPDENet offers enhanced generalizability."}
+{"idx": 1, "title": "Multi-scale time-stepping of Partial Differential Equations with ...", "date": "", "ddg_snippet": "The equation was solved using the stream-function formulation with a pseudospectral method, the time-stepping was done using the Crank-Nicolson scheme with a time -step of 1 e 4, and the solution was recorded every t = 1 time units.", "subpage_snippet": "", "source": "www.sciencedirect.com", "link": "https://www.sciencedirect.com/science/article/pii/S0045782524002391", "content": "The equation was solved using the stream-function formulation with a pseudospectral method, the time-stepping was done using the Crank-Nicolson scheme with a time -step of 1 e 4, and the solution was recorded every t = 1 time units."}
+{"idx": 2, "title": "PDF Multi-Scale Message Passing Neural PDE Solvers", "date": "", "ddg_snippet": "We propose a novel multi-scale message passing neural network algorithm for learning the solutions of time -dependent PDEs . Our algorithm possesses both temporal and spatial multi-scale resolution features by incorporating multi-scale sequence models and graph gating modules in the encoder and processor, respec-tively. Benchmark numerical experiments are presented to demonstrate that the ...", "subpage_snippet": "", "source": "www.sam.math.ethz.ch", "link": "https://www.sam.math.ethz.ch/sam_reports/reports_final/reports2023/2023-14.pdf", "content": "We propose a novel multi-scale message passing neural network algorithm for learning the solutions of time -dependent PDEs . Our algorithm possesses both temporal and spatial multi-scale resolution features by incorporating multi-scale sequence models and graph gating modules in the encoder and processor, respec-tively. Benchmark numerical experiments are presented to demonstrate that the ..."}
+{"idx": 3, "title": "Neural network-based time stepping scheme for multiscale partial ...", "date": "", "ddg_snippet": "In this paper, we demonstrate the application of deep learning techniques to derive an accurate time stepping scheme for multiscale systems. The proposed technique can capture the multiscale PDE behavior with high accuracy while being computationally tractable.", "subpage_snippet": "", "source": "ieeexplore.ieee.org", "link": "https://ieeexplore.ieee.org/document/10455439", "content": "In this paper, we demonstrate the application of deep learning techniques to derive an accurate time stepping scheme for multiscale systems. The proposed technique can capture the multiscale PDE behavior with high accuracy while being computationally tractable."}
+{"idx": 4, "title": "ICML Poster MultiPDENet: PDE-embedded Learning with Multi-time-stepping ...", "date": "", "ddg_snippet": "To this end, we propose a PDE -embedded network with multiscale time stepping (MultiPDENet), which fuses the scheme of numerical methods and machine learning, for accelerated simulation of flows.", "subpage_snippet": "", "source": "icml.cc", "link": "https://icml.cc/virtual/2025/poster/46029", "content": "To this end, we propose a PDE -embedded network with multiscale time stepping (MultiPDENet), which fuses the scheme of numerical methods and machine learning, for accelerated simulation of flows."}
+{"idx": 5, "title": "TSONN: Time-stepping-oriented neural network for solving partial ...", "date": "", "ddg_snippet": "Deep neural networks (DNNs), especially physics-informed neural networks (PINNs), have recently become a new popular method for solving forward and inverse problems governed by partial differential equations ( PDEs ). However, these methods still face challenges in achieving stable training and obtaining correct results in many problems, since minimizing PDE residuals with PDE -based soft ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2310.16491", "content": "Deep neural networks (DNNs), especially physics-informed neural networks (PINNs), have recently become a new popular method for solving forward and inverse problems governed by partial differential equations ( PDEs ). However, these methods still face challenges in achieving stable training and obtaining correct results in many problems, since minimizing PDE residuals with PDE -based soft ..."}
+{"idx": 6, "title": "PDE-constrained Learning with Multi-time-stepping for Accelerated...", "date": "", "ddg_snippet": "To this end, we propose a PDE -embedded network with multiscale time stepping (MultiPDENet), which fuses the scheme of numerical methods and machine learning, for accelerated simulation of fluid flows.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=stcN89QGfL", "content": "To this end, we propose a PDE -embedded network with multiscale time stepping (MultiPDENet), which fuses the scheme of numerical methods and machine learning, for accelerated simulation of fluid flows."}
+{"idx": 7, "title": "Blending neural operators and relaxation methods in PDE ... - Nature", "date": "", "ddg_snippet": "Neural -network-based solvers for partial differential equations ( PDEs ) suffer from difficulties tackling high-frequency modes when learning complex functions, whereas for classical solvers it is ...", "subpage_snippet": "", "source": "www.nature.com", "link": "https://www.nature.com/articles/s42256-024-00910-x", "content": "Neural -network-based solvers for partial differential equations ( PDEs ) suffer from difficulties tackling high-frequency modes when learning complex functions, whereas for classical solvers it is ..."}
+{"idx": 8, "title": "Multiscale Neural Operators for Solving Time-Independent PDEs", "date": "", "ddg_snippet": "TL;DR: We study how to solve time -independent Partial Differential Equations on large meshes and introduce a novel graph rewiring technique for this. Multiscale Neural Operators for Solving Time -Independent PDEs by Winfried Ripken 1 *, Lisa Coiffard 1 *, Felix Pieper 1 * and Sebastian Dziadzio 2. 1 Merantix Momentum, 2 Tübingen AI Center. (*) equal contribution.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/merantix-momentum/multiscale-pde-operators", "content": "TL;DR: We study how to solve time -independent Partial Differential Equations on large meshes and introduce a novel graph rewiring technique for this. Multiscale Neural Operators for Solving Time -Independent PDEs by Winfried Ripken 1 *, Lisa Coiffard 1 *, Felix Pieper 1 * and Sebastian Dziadzio 2. 1 Merantix Momentum, 2 Tübingen AI Center. (*) equal contribution."}
+{"idx": 9, "title": "A neural network-based PDE solving algorithm with high precision", "date": "", "ddg_snippet": "The consumption of solving large-scale linear equations is one of the most critical issues in numerical computation. An innovative method is introduced in this study to solve linear equations ...", "subpage_snippet": "", "source": "www.nature.com", "link": "https://www.nature.com/articles/s41598-023-31236-0", "content": "The consumption of solving large-scale linear equations is one of the most critical issues in numerical computation. An innovative method is introduced in this study to solve linear equations ..."}
diff --git a/data/sampled_jsons/optimal_LP_solution_online_matching_probability_1n_common_type.jsonl b/data/sampled_jsons/optimal_LP_solution_online_matching_probability_1n_common_type.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..f8cd5908c9f5638ba323047f00b7ee4eb1281070
--- /dev/null
+++ b/data/sampled_jsons/optimal_LP_solution_online_matching_probability_1n_common_type.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Promoting Fairness Among Dynamic Agents in Online ...", "date": "", "ddg_snippet": "9 Dec 2024 — In this paper, we study online matching problems where performance is instead determined by the fairness in service provided to different online ...", "subpage_snippet": "", "source": "neurips.cc", "link": "https://neurips.cc/virtual/2024/poster/96945", "content": "9 Dec 2024 — In this paper, we study online matching problems where performance is instead determined by the fairness in service provided to different online ..."}
+{"idx": 1, "title": "A Dynamic Near-Optimal Algorithm for Online Linear ...", "date": "", "ddg_snippet": "by S Agrawal · 2009 · Cited by 391 — A natural optimization model that formulates many online resource allocation and revenue management problems is the online linear program (LP) ...", "subpage_snippet": "", "source": "web.stanford.edu", "link": "https://web.stanford.edu/~yyye/onlinelp.pdf", "content": "by S Agrawal · 2009 · Cited by 391 — A natural optimization model that formulates many online resource allocation and revenue management problems is the online linear program (LP) ..."}
+{"idx": 2, "title": "A Dynamic Near-Optimal Algorithm for Online Linear ...", "date": "", "ddg_snippet": "by S Agrawal · Cited by 391 — In this paper, we propose a near- optimal algorithm for this general class of online problems under the assumptions of random order of arrival and some mild ... 30 pages", "subpage_snippet": "", "source": "web.stanford.edu", "link": "https://web.stanford.edu/~yyye/ORonlinelp_final.pdf", "content": "by S Agrawal · Cited by 391 — In this paper, we propose a near- optimal algorithm for this general class of online problems under the assumptions of random order of arrival and some mild ... 30 pages"}
+{"idx": 3, "title": "Online Stochastic Matching: New Algorithms and Bounds", "date": "", "ddg_snippet": "by B Brubach · 2016 · Cited by 41 — We use the optimal solution to this LP to guide our online actions. In most cases, we use various modifications of dependent randomized rounding ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/1606.06395", "content": "by B Brubach · 2016 · Cited by 41 — We use the optimal solution to this LP to guide our online actions. In most cases, we use various modifications of dependent randomized rounding ..."}
+{"idx": 4, "title": "Promoting Fairness Among Dynamic Agents in Online- ...", "date": "", "ddg_snippet": "by W Ma · 2024 · Cited by 1 — 1 Solve LP (10) to get an optimal solution {x∗ ij }. 2 Let an online agent (of type ) j arrive at time t. 3 Sample a neighbor i ∈ Nj with probability x∗. 30 pages", "subpage_snippet": "", "source": "proceedings.neurips.cc", "link": "https://proceedings.neurips.cc/paper_files/paper/2024/file/959f70ee50044bed305e48e3484005a7-Paper-Conference.pdf", "content": "by W Ma · 2024 · Cited by 1 — 1 Solve LP (10) to get an optimal solution {x∗ ij }. 2 Let an online agent (of type ) j arrive at time t. 3 Sample a neighbor i ∈ Nj with probability x∗. 30 pages"}
+{"idx": 5, "title": "A Dynamic Near-Optimal Algorithm for Online Linear ...", "date": "", "ddg_snippet": "by S Agrawal · 2014 · Cited by 391 — We also present a worst case example showing that the performance of our algorithm is near optimal . Subject classifications: online algorithms; ...", "subpage_snippet": "", "source": "www.jstor.org", "link": "https://www.jstor.org/stable/24540667", "content": "by S Agrawal · 2014 · Cited by 391 — We also present a worst case example showing that the performance of our algorithm is near optimal . Subject classifications: online algorithms; ..."}
+{"idx": 6, "title": "Online Stochastic Matching: New Algorithms with Better ...", "date": "", "ddg_snippet": "by P Jaillet · Cited by 200 — From now on, we will compare the online algorithm with the optimal solution of LP (2) instead of with the offline solution , because the former one is much ... 30 pages", "subpage_snippet": "", "source": "web.mit.edu", "link": "https://web.mit.edu/jaillet/www/general/matching_pj_xl-final-mor-6-13.pdf", "content": "by P Jaillet · Cited by 200 — From now on, we will compare the online algorithm with the optimal solution of LP (2) instead of with the offline solution , because the former one is much ... 30 pages"}
+{"idx": 7, "title": "Randomized Rounding Approaches to Online Allocation, ...", "date": "", "ddg_snippet": "by W Ma · 2024 · Cited by 1 — E[Di] = T/n + 1 − 1 / n ≤ 2T/n. Therefore, setting yi = 1 for all i ∈ [n] forms an optimal solution to LP (16), with objective value n( 1 / n ) = 1.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2407.20419", "content": "by W Ma · 2024 · Cited by 1 — E[Di] = T/n + 1 − 1 / n ≤ 2T/n. Therefore, setting yi = 1 for all i ∈ [n] forms an optimal solution to LP (16), with objective value n( 1 / n ) = 1."}
+{"idx": 8, "title": "Efficient Joint Object Matching via Linear Programming", "date": "", "ddg_snippet": "by A De Rosa · 2021 · Cited by 7 — We say that an optimization algorithm recovers the ground truth, whenever its unique optimal solution coincides with the ground truth. In the ...", "subpage_snippet": "", "source": "optimization-online.org", "link": "https://optimization-online.org/wp-content/uploads/2021/08/8568.pdf", "content": "by A De Rosa · 2021 · Cited by 7 — We say that an optimization algorithm recovers the ground truth, whenever its unique optimal solution coincides with the ground truth. In the ..."}
+{"idx": 9, "title": "Online Stochastic Matching: New Algorithms and Bounds", "date": "", "ddg_snippet": "by B Brubach · 2020 · Cited by 41 — We use the optimal solution to this LP to guide our online actions. ... (via Monte-Carlo simulation) the probability of matching an edge e in the offline optimal ... 47 pages", "subpage_snippet": "", "source": "www.cs.umd.edu", "link": "https://www.cs.umd.edu/~srin/PDF/2020/online-stoch-match.pdf", "content": "by B Brubach · 2020 · Cited by 41 — We use the optimal solution to this LP to guide our online actions. ... (via Monte-Carlo simulation) the probability of matching an edge e in the offline optimal ... 47 pages"}
diff --git a/data/sampled_jsons/optimism-based_exploration_continuous_MDPs_neural_networks.jsonl b/data/sampled_jsons/optimism-based_exploration_continuous_MDPs_neural_networks.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..6f73de6d8c94e231c2ce9793ef4157d9a0f1ee6b
--- /dev/null
+++ b/data/sampled_jsons/optimism-based_exploration_continuous_MDPs_neural_networks.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Optimistic Initialization for Exploration in Continuous Control", "date": "", "ddg_snippet": "Optimistic initialization is foundational to reinforcement learning in discrete MDPs ; in this paper we develop tech-niques that successfully apply this concept to deep neural networks used to solve continuous control problems.", "subpage_snippet": "", "source": "irl.cs.brown.edu", "link": "http://irl.cs.brown.edu/pubs/optinit_explore_control.pdf", "content": "Optimistic initialization is foundational to reinforcement learning in discrete MDPs ; in this paper we develop tech-niques that successfully apply this concept to deep neural networks used to solve continuous control problems."}
+{"idx": 1, "title": "Bayesian Optimistic Optimization: Optimistic Exploration for ...", "date": "", "ddg_snippet": "H-UCRL was devoted to resolving the intractability of model- based optimistic exploration for general models [26]. It proposes to convert the joint optimization of model and policy into a hallucinated control problem.", "subpage_snippet": "", "source": "proceedings.neurips.cc", "link": "https://proceedings.neurips.cc/paper_files/paper/2022/file/5bcb807ae43ad0851a6ba6162a866404-Paper-Conference.pdf", "content": "H-UCRL was devoted to resolving the intractability of model- based optimistic exploration for general models [26]. It proposes to convert the joint optimization of model and policy into a hallucinated control problem."}
+{"idx": 2, "title": "OPTIMISTIC EXPLORATION WITH LEARNED FEATURES PROVABLY SOLVES ...", "date": "", "ddg_snippet": "As we mentioned in §3, the transition of MDPs with neural dynamics can be written as an energy- based model and admits the Gaussian RBF kernel. To exploit the kernel structure in the transition, we explore the environment using the exploration bonus induced by the Gaussian RBF kernel and the feature maps learned from the data, which is ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=9kBCMNb5mc", "content": "As we mentioned in §3, the transition of MDPs with neural dynamics can be written as an energy- based model and admits the Gaussian RBF kernel. To exploit the kernel structure in the transition, we explore the environment using the exploration bonus induced by the Gaussian RBF kernel and the feature maps learned from the data, which is ..."}
+{"idx": 3, "title": "Optimistic Initialization for Exploration in Continuous ...", "date": "", "ddg_snippet": "Jun 28, 2022 · Here we use recent advances in training deep neural networks to develop a novel artificial agent, termed a deep Q- network , that can learn successful policies directly from high-dimensional sensory ...", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/publication/361744301_Optimistic_Initialization_for_Exploration_in_Continuous_Control", "content": "Jun 28, 2022 · Here we use recent advances in training deep neural networks to develop a novel artificial agent, termed a deep Q- network , that can learn successful policies directly from high-dimensional sensory ..."}
+{"idx": 4, "title": "Beyond optimism | Proceedings of the 38th International ...", "date": "", "ddg_snippet": "Jun 5, 2025 · To improve exploration and reward discovery, popular algorithms rely on optimism. But what if sometimes rewards are unobservable, e.g., situations of partial monitoring in bandits and the recent formalism of monitored Markov decision process? In this case, optimism can lead to suboptimal behavior that does not explore further to collapse ...", "subpage_snippet": "", "source": "dl.acm.org", "link": "https://dl.acm.org/doi/abs/10.5555/3737916.3740005", "content": "Jun 5, 2025 · To improve exploration and reward discovery, popular algorithms rely on optimism. But what if sometimes rewards are unobservable, e.g., situations of partial monitoring in bandits and the recent formalism of monitored Markov decision process? In this case, optimism can lead to suboptimal behavior that does not explore further to collapse ..."}
+{"idx": 5, "title": "Optimism-Driven Exploration for Nonlinear Systems", "date": "", "ddg_snippet": "Model- based RL reduces the required interaction time by learning a model of the dynamics. Although many methods have been proposed for model- based RL with efficient exploration in discrete MDPs , data-efficient model- based RL in continuous systems remains a challenging problem despite substantial recent advances [3], [4], [5].", "subpage_snippet": "", "source": "people.eecs.berkeley.edu", "link": "https://people.eecs.berkeley.edu/~pabbeel/papers/2015-ICRA-exploration-nonlinear-systems.pdf", "content": "Model- based RL reduces the required interaction time by learning a model of the dynamics. Although many methods have been proposed for model- based RL with efficient exploration in discrete MDPs , data-efficient model- based RL in continuous systems remains a challenging problem despite substantial recent advances [3], [4], [5]."}
+{"idx": 6, "title": "Principled Exploration via Optimistic Bootstrapping and ...", "date": "", "ddg_snippet": "Devising efficient exploration algorithms thus becomes an attractive topic in recent years of RL research. The theo-retical achievements in RL offer various provably efficient exploration methods in tabular and linear Markov Decision Processes ( MDPs ) based on the fundamental value iteration algorithm Least-Squares Value Iteration (LSVI).", "subpage_snippet": "", "source": "proceedings.mlr.press", "link": "https://proceedings.mlr.press/v139/bai21d/bai21d.pdf", "content": "Devising efficient exploration algorithms thus becomes an attractive topic in recent years of RL research. The theo-retical achievements in RL offer various provably efficient exploration methods in tabular and linear Markov Decision Processes ( MDPs ) based on the fundamental value iteration algorithm Least-Squares Value Iteration (LSVI)."}
+{"idx": 7, "title": "Exploration in Reinforcement Learning: Beyond Finite State-Spaces", "date": "", "ddg_snippet": "A Kernel- Based Approach to Exploration in Continuous MDPs .", "subpage_snippet": "", "source": "theses.hal.science", "link": "https://theses.hal.science/tel-03892761v2/file/These_DARWICHE_DOMINGUES_Omar.pdf", "content": "A Kernel- Based Approach to Exploration in Continuous MDPs ."}
+{"idx": 8, "title": "VIME: Variational Information Maximizing Exploration", "date": "", "ddg_snippet": "Some of these algorithms are based on the principle of optimism under uncertainty: E3 [3], R-Max [4], UCRL [30]. An alternative approach is Bayesian reinforcement learning methods, which maintain a distribution over possible MDPs [1, 17, 23, 31].", "subpage_snippet": "", "source": "proceedings.neurips.cc", "link": "https://proceedings.neurips.cc/paper/2016/file/abd815286ba1007abfbb8415b83ae2cf-Paper.pdf", "content": "Some of these algorithms are based on the principle of optimism under uncertainty: E3 [3], R-Max [4], UCRL [30]. An alternative approach is Bayesian reinforcement learning methods, which maintain a distribution over possible MDPs [1, 17, 23, 31]."}
+{"idx": 9, "title": "MobILE : Model- Based Imitation Learning From", "date": "", "ddg_snippet": "incentivize exploration . We utilize an neural network ensemble, where each model is trained on Dt (via SGD on squared loss Eq. 2) with different initialization. Neural network dynamics for model- based deep reinforcement learning with model-free ne-tuning.", "subpage_snippet": "", "source": "papers.nips.cc", "link": "https://papers.nips.cc/paper/2021/file/f06048518ff8de2035363e00710c6a1d-Paper.pdf", "content": "incentivize exploration . We utilize an neural network ensemble, where each model is trained on Dt (via SGD on squared loss Eq. 2) with different initialization. Neural network dynamics for model- based deep reinforcement learning with model-free ne-tuning."}
diff --git a/data/sampled_jsons/probability_of_necessity_Galhotra_Halpern.jsonl b/data/sampled_jsons/probability_of_necessity_Galhotra_Halpern.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..9652e2a1e56a04d8c22caeb31bf449a24107c682
--- /dev/null
+++ b/data/sampled_jsons/probability_of_necessity_Galhotra_Halpern.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Local Explanations via Necessity and Sufciency", "date": "", "ddg_snippet": "Necessity and sufciency are the building blocks of all successful explanations.[2020a] build on Halpern [2016]’s denitions of necessity and sufciency to critique popular XAI tools, proposing a new feature attribution measure with some pur-ported advantages.", "subpage_snippet": "", "source": "philarchive.org", "link": "https://philarchive.org/archive/WATLEVv1", "content": "Necessity and sufciency are the building blocks of all successful explanations.[2020a] build on Halpern [2016]’s denitions of necessity and sufciency to critique popular XAI tools, proposing a new feature attribution measure with some pur-ported advantages."}
+{"idx": 1, "title": "Local Explanations via Necessity and Sufciency", "date": "", "ddg_snippet": "Necessity and sufciency have a long philosophical tradi-tion [Mackie, 1965; Lewis, 1973; Halpern and Pearl, 2005], spanning logical, probabilistic, and causal variants.Denition 3 ( Probability of Necessity ). The probability that c is a necessary factor for outcome y is given by", "subpage_snippet": "", "source": "discovery.ucl.ac.uk", "link": "https://discovery.ucl.ac.uk/id/eprint/10144541/1/watson21a.pdf", "content": "Necessity and sufciency have a long philosophical tradi-tion [Mackie, 1965; Lewis, 1973; Halpern and Pearl, 2005], spanning logical, probabilistic, and causal variants.Denition 3 ( Probability of Necessity ). The probability that c is a necessary factor for outcome y is given by"}
+{"idx": 2, "title": "Local Explanations via Necessity and Sufficiency: Unifying Theory and...", "date": "", "ddg_snippet": "Necessity and sufficiency are the building blocks of all successful explanations.of independent interest. Thus, while necessity is both the converse and inverse of sufficiency in propositional logic, the two formulations come apart in probability calculus.", "subpage_snippet": "", "source": "link.springer.com", "link": "https://link.springer.com/article/10.1007/s11023-022-09598-7", "content": "Necessity and sufficiency are the building blocks of all successful explanations.of independent interest. Thus, while necessity is both the converse and inverse of sufficiency in propositional logic, the two formulations come apart in probability calculus."}
+{"idx": 3, "title": "causality - How to understand probability of Necessity ...", "date": "", "ddg_snippet": "We see that while the observation ERR is negative (-13) giving the impression that the drug is actually preventing deaths , the bias-correction term (+14) rectifies this impression and set the probability of necessity (PN) to unity. Moreover, since the lower bound of Eq.", "subpage_snippet": "", "source": "stats.stackexchange.com", "link": "https://stats.stackexchange.com/questions/445886/how-to-understand-probability-of-necessity-pn-≥-100-as-in-this-example-from", "content": "We see that while the observation ERR is negative (-13) giving the impression that the drug is actually preventing deaths , the bias-correction term (+14) rectifies this impression and set the probability of necessity (PN) to unity. Moreover, since the lower bound of Eq."}
+{"idx": 4, "title": "Identifying and bounding the probability of necessity for... | CoLab", "date": "", "ddg_snippet": "In backward-looking causal inference, the probability of necessity measures the probability that a certain event is caused by the treatment given the observed treatment and outcome. Most existing results focus on binary outcomes.", "subpage_snippet": "", "source": "colab.ws", "link": "https://colab.ws/articles/10.1093/biomet/asaf049", "content": "In backward-looking causal inference, the probability of necessity measures the probability that a certain event is caused by the treatment given the observed treatment and outcome. Most existing results focus on binary outcomes."}
+{"idx": 5, "title": "Feature Attribution with Necessity and Sufficiencyvia Dual-stage...", "date": "", "ddg_snippet": "( Probability of Necessity and Sufficiency, PNS).PNS is the weighted sum of PN and PS, each multiplied by the probability of its corresponding condition. PNS mea-sures the probability that As,b is a necessary and sufficient cause for Bs,c (as the “if and only if” of the relationship).", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=J6prHJsIlf", "content": "( Probability of Necessity and Sufficiency, PNS).PNS is the weighted sum of PN and PS, each multiplied by the probability of its corresponding condition. PNS mea-sures the probability that As,b is a necessary and sufficient cause for Bs,c (as the “if and only if” of the relationship)."}
+{"idx": 6, "title": "The probability of necessity in causal analysis - YouTube", "date": "", "ddg_snippet": "О сервисе Прессе Авторские права Связаться с нами Авторам Рекламодателям...", "subpage_snippet": "", "source": "www.youtube.com", "link": "https://www.youtube.com/watch?v=Sl6zf_gYg-A", "content": "О сервисе Прессе Авторские права Связаться с нами Авторам Рекламодателям..."}
+{"idx": 7, "title": "Learning Probabilities of Causation from Finite Population Data", "date": "", "ddg_snippet": "Definition 1 ( Probability of necessity (PN)). Let X and Y be two binary variables in a causal model M , let x and y stand for the propositions X = true and Y = true, respectively, and x′ and y′ for their complements. The probability of necessity is defined as the expression.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2505.17133", "content": "Definition 1 ( Probability of necessity (PN)). Let X and Y be two binary variables in a causal model M , let x and y stand for the propositions X = true and Y = true, respectively, and x′ and y′ for their complements. The probability of necessity is defined as the expression."}
+{"idx": 8, "title": "Probabilities Of Causation: Three Counterfactual Interpretations And...", "date": "", "ddg_snippet": "The probability of necessity is dened as the expression.Thus, mirroring the necessity of x (as measured by PN), PS gives the probability that setting x would produce y in a situation where x and y are in fact absent.", "subpage_snippet": "", "source": "ftp.cs.ucla.edu", "link": "https://ftp.cs.ucla.edu/pub/stat_ser/r260-reprint.pdf", "content": "The probability of necessity is dened as the expression.Thus, mirroring the necessity of x (as measured by PN), PS gives the probability that setting x would produce y in a situation where x and y are in fact absent."}
+{"idx": 9, "title": "Lecture-15", "date": "", "ddg_snippet": "– Counterfactual c auses – Actual causes ( Halpern & Pearl). • Measuring causality. – Responsibility – Probability of necessity , Probability of sufficiency.", "subpage_snippet": "", "source": "courses.cs.duke.edu", "link": "https://courses.cs.duke.edu/fall15/cps296.6/Lectures/Lecture-15.pdf", "content": "– Counterfactual c auses – Actual causes ( Halpern & Pearl). • Measuring causality. – Responsibility – Probability of necessity , Probability of sufficiency."}
diff --git a/data/sampled_jsons/reverse_cross_entropy_RCE_symmetric_cross_entropy_loss_formula.jsonl b/data/sampled_jsons/reverse_cross_entropy_RCE_symmetric_cross_entropy_loss_formula.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..0eae63e3889060054a7250a80cb200e5f36addb9
--- /dev/null
+++ b/data/sampled_jsons/reverse_cross_entropy_RCE_symmetric_cross_entropy_loss_formula.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Learning to Rematch Mismatched Pairs for Robust Cross-Modal", "date": "", "ddg_snippet": "... of cross -modal pairs, which generally follow two research lines: 1) Global Alignment focuses on learning the correspondence between whole cross -modal ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2403.05105v1", "content": "... of cross -modal pairs, which generally follow two research lines: 1) Global Alignment focuses on learning the correspondence between whole cross -modal ..."}
+{"idx": 1, "title": "Symmetric Cross Entropy for Robust Learning with Noisy Labels", "date": "", "ddg_snippet": "Inspired by the symmetric KL-divergence, we propose the approach of \\textbf { Symmetric cross entropy Learning} (SL), boosting CE symmetrically with a noise robust counterpart Reverse Cross Entropy ( RCE ). Our proposed SL approach simultaneously addresses both the under learning and overfitting problem of CE in the presence of noisy labels.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/1908.06112", "content": "Inspired by the symmetric KL-divergence, we propose the approach of \\textbf { Symmetric cross entropy Learning} (SL), boosting CE symmetrically with a noise robust counterpart Reverse Cross Entropy ( RCE ). Our proposed SL approach simultaneously addresses both the under learning and overfitting problem of CE in the presence of noisy labels."}
+{"idx": 2, "title": "PDF Can Cross Entropy Loss Be Robust to Label Noise?", "date": "", "ddg_snippet": "Symmetric Cross Entropy (SCE) [Wang et al., 2019] combines CCE and Reverse Cross Entropy ( RCE , which is equivalent to MAE) by tuning the regularization parameters.", "subpage_snippet": "", "source": "personal.ntu.edu.sg", "link": "https://personal.ntu.edu.sg/boan/papers/IJCAI20_Entropy.pdf", "content": "Symmetric Cross Entropy (SCE) [Wang et al., 2019] combines CCE and Reverse Cross Entropy ( RCE , which is equivalent to MAE) by tuning the regularization parameters."}
+{"idx": 3, "title": "Reverse Cross Entropy for Adversarial Detection (NeurIPS 2018)", "date": "", "ddg_snippet": "Reverse Cross Entropy Training Reverse Cross Entropy Training ( RCE ) is a novel training method, which can learn more distinguished feature representations for detecting adversarial examples.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/P2333/Reverse-Cross-Entropy", "content": "Reverse Cross Entropy Training Reverse Cross Entropy Training ( RCE ) is a novel training method, which can learn more distinguished feature representations for detecting adversarial examples."}
+{"idx": 4, "title": "PDF Symmetric Cross Entropy for Robust Learning with Noisy Labels", "date": "", "ddg_snippet": "Inspired by the symmetric KL-divergence, we pro-pose the approach of Symmetric cross entropy Learning (SL), boosting CE symmetrically with a noise robust coun-terpart Reverse Cross Entropy ( RCE ). Our proposed SL ap-proach simultaneously addresses both the under learning and overfitting problem of CE in the presence of noisy la-bels.", "subpage_snippet": "", "source": "openaccess.thecvf.com", "link": "https://openaccess.thecvf.com/content_ICCV_2019/papers/Wang_Symmetric_Cross_Entropy_for_Robust_Learning_With_Noisy_Labels_ICCV_2019_paper.pdf", "content": "Inspired by the symmetric KL-divergence, we pro-pose the approach of Symmetric cross entropy Learning (SL), boosting CE symmetrically with a noise robust coun-terpart Reverse Cross Entropy ( RCE ). Our proposed SL ap-proach simultaneously addresses both the under learning and overfitting problem of CE in the presence of noisy la-bels."}
+{"idx": 5, "title": "What Is Cross-Entropy Loss Function? - GeeksforGeeks", "date": "", "ddg_snippet": "Cross - entropy loss is a way to measure how close a model's predictions are to the correct answers in classification problems. It helps train models to make more confident and accurate predictions by rewarding correct answers and penalizing wrong ones. This makes it a key part of building reliable machine learning classifiers.", "subpage_snippet": "", "source": "www.geeksforgeeks.org", "link": "https://www.geeksforgeeks.org/machine-learning/what-is-cross-entropy-loss-function/", "content": "Cross - entropy loss is a way to measure how close a model's predictions are to the correct answers in classification problems. It helps train models to make more confident and accurate predictions by rewarding correct answers and penalizing wrong ones. This makes it a key part of building reliable machine learning classifiers."}
+{"idx": 6, "title": "Hierarchical symmetric cross entropy for distant supervised relation ...", "date": "", "ddg_snippet": "The Symmetric Cross Entropy (SCE) loss function combines the Cross Entropy (CE) and Reverse Cross Entropy ( RCE ) to improve the robustness of the model under noisy labels.", "subpage_snippet": "", "source": "link.springer.com", "link": "https://link.springer.com/article/10.1007/s10489-024-05798-z", "content": "The Symmetric Cross Entropy (SCE) loss function combines the Cross Entropy (CE) and Reverse Cross Entropy ( RCE ) to improve the robustness of the model under noisy labels."}
+{"idx": 7, "title": "Robust CNN model using LDAM DRW and RCE Loss functions", "date": "", "ddg_snippet": "Paper proposes introduction of a noise robust Reverse Cross Entropy term to reduce overfitting of easy classes and under learning of hard classes Ref [28, 1] state that DNNs first memorize data for clean labels, then for noisy labels. Eg. in ref [13,19]. Cross Entropy causes class-biased learning ie. Easy classes converge faster than hard classes.", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/AninditaDeb/LDAM-DRW-and-RCE", "content": "Paper proposes introduction of a noise robust Reverse Cross Entropy term to reduce overfitting of easy classes and under learning of hard classes Ref [28, 1] state that DNNs first memorize data for clean labels, then for noisy labels. Eg. in ref [13,19]. Cross Entropy causes class-biased learning ie. Easy classes converge faster than hard classes."}
+{"idx": 8, "title": "Symmetric Reinforcement Learning Loss for Robust Learning on...", "date": "", "ddg_snippet": "In this work, we focus on RL algorithms that share learning difficulties with cross - entropy loss , especially for low-probability predictions. To enhance stability, we adapt reverse cross - entropy ( RCE ) from supervised learning for noisy data, defining a symmetric RL loss . We demonstrate performance improvements across various tasks and scales.", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/forum?id=YjBrt82S3v", "content": "In this work, we focus on RL algorithms that share learning difficulties with cross - entropy loss , especially for low-probability predictions. To enhance stability, we adapt reverse cross - entropy ( RCE ) from supervised learning for noisy data, defining a symmetric RL loss . We demonstrate performance improvements across various tasks and scales."}
+{"idx": 9, "title": "Outlier-Robust Training of Machine Learning Models", "date": "", "ddg_snippet": "... italic_i end_POSTSUBSCRIPT ( bold_italic_w ) is the cross - entropy loss , minimizing the squared cross - entropy loss does not make an equal sense.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2501.00265v1", "content": "... italic_i end_POSTSUBSCRIPT ( bold_italic_w ) is the cross - entropy loss , minimizing the squared cross - entropy loss does not make an equal sense."}
diff --git a/data/sampled_jsons/simultaneously_minimizing_loss_value_and_loss_sharpness_SAM_abstract.jsonl b/data/sampled_jsons/simultaneously_minimizing_loss_value_and_loss_sharpness_SAM_abstract.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..4bad43f985e2cecb5319574b6180ba01899f12c3
--- /dev/null
+++ b/data/sampled_jsons/simultaneously_minimizing_loss_value_and_loss_sharpness_SAM_abstract.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "[2010.01412] Sharpness-Aware Minimization for Efficiently Improving ...", "date": "", "ddg_snippet": "Indeed, optimizing only the training loss value , as is commonly done, can easily lead to suboptimal model quality. Motivated by prior work connecting the geometry of the loss landscape and generalization, we introduce a novel, effective procedure for instead simultaneously minimizing loss value and loss sharpness .", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2010.01412", "content": "Indeed, optimizing only the training loss value , as is commonly done, can easily lead to suboptimal model quality. Motivated by prior work connecting the geometry of the loss landscape and generalization, we introduce a novel, effective procedure for instead simultaneously minimizing loss value and loss sharpness ."}
+{"idx": 1, "title": "Random Sharpness-Aware Minimization - NIPS", "date": "", "ddg_snippet": "In particular, a minimax optimization objective is defined to find the maximum loss value centered on the weight, out of the purpose of simultaneously minimizing loss value and loss sharpness . For the sake of simplicity, SAM applies one-step gradient ascent to approximate the solution of the inner maximization.", "subpage_snippet": "", "source": "papers.nips.cc", "link": "https://papers.nips.cc/paper_files/paper/2022/hash/9b79416c0dc4b09feaa169ed5cdd63d4-Abstract-Conference.html", "content": "In particular, a minimax optimization objective is defined to find the maximum loss value centered on the weight, out of the purpose of simultaneously minimizing loss value and loss sharpness . For the sake of simplicity, SAM applies one-step gradient ascent to approximate the solution of the inner maximization."}
+{"idx": 2, "title": "PDF Random Sharpness-Aware Minimization - NeurIPS", "date": "", "ddg_snippet": "Abstract Currently, Sharpness -Aware Minimization ( SAM ) is proposed to seek the param-eters that lie in a flat region to improve the generalization when training neural networks. In particular, a minimax optimization objective is defined to find the maximum loss value centered on the weight, out of the purpose of simultaneously minimizing loss value and loss sharpness . For the sake of ...", "subpage_snippet": "", "source": "proceedings.neurips.cc", "link": "https://proceedings.neurips.cc/paper_files/paper/2022/file/9b79416c0dc4b09feaa169ed5cdd63d4-Paper-Conference.pdf", "content": "Abstract Currently, Sharpness -Aware Minimization ( SAM ) is proposed to seek the param-eters that lie in a flat region to improve the generalization when training neural networks. In particular, a minimax optimization objective is defined to find the maximum loss value centered on the weight, out of the purpose of simultaneously minimizing loss value and loss sharpness . For the sake of ..."}
+{"idx": 3, "title": "ICLR 2021 Sharpness-aware Minimization for Efficiently Improving ...", "date": "", "ddg_snippet": "Motivated by the connection between geometry of the loss landscape and generalization---including a generalization bound that we prove here---we introduce a novel, effective procedure for instead simultaneously minimizing loss value and loss sharpness .", "subpage_snippet": "", "source": "iclr.cc", "link": "https://iclr.cc/virtual/2021/spotlight/3497", "content": "Motivated by the connection between geometry of the loss landscape and generalization---including a generalization bound that we prove here---we introduce a novel, effective procedure for instead simultaneously minimizing loss value and loss sharpness ."}
+{"idx": 4, "title": "GitHub - davda54/sam: SAM: Sharpness-Aware Minimization (PyTorch)", "date": "", "ddg_snippet": "SAM simultaneously minimizes loss value and loss sharpness . In particular, it seeks parameters that lie in neighborhoods having uniformly low loss . SAM improves model generalization and yields SoTA performance for several datasets. Additionally, it provides robustness to label noise on par with that provided by SoTA procedures that specifically target learning with noisy labels. This is an ...", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/davda54/sam", "content": "SAM simultaneously minimizes loss value and loss sharpness . In particular, it seeks parameters that lie in neighborhoods having uniformly low loss . SAM improves model generalization and yields SoTA performance for several datasets. Additionally, it provides robustness to label noise on par with that provided by SoTA procedures that specifically target learning with noisy labels. This is an ..."}
+{"idx": 5, "title": "Paper page - Sharpness-Aware Minimization for Efficiently Improving ...", "date": "", "ddg_snippet": "Abstract Sharpness -Aware Minimization ( SAM ) improves model generalization by simultaneously minimizing loss value and sharpness , achieving state-of-the-art performance and label noise robustness.", "subpage_snippet": "", "source": "huggingface.co", "link": "https://huggingface.co/papers/2010.01412", "content": "Abstract Sharpness -Aware Minimization ( SAM ) improves model generalization by simultaneously minimizing loss value and sharpness , achieving state-of-the-art performance and label noise robustness."}
+{"idx": 6, "title": "Improving Sharpness-Aware Minimization Using Label Smoothing and ...", "date": "", "ddg_snippet": "Recent advances in learning algorithms have identified loss surface sharpness as an effective metric for reducing the generalization gap. Building on this principle, Sharpness -Aware Minimization ( SAM ) was introduced to improve model generalization and has achieved state-of-the-art performance through its weight perturbation and updating steps. However, SAM's perturbation is based solely on ...", "subpage_snippet": "", "source": "ieeexplore.ieee.org", "link": "https://ieeexplore.ieee.org/abstract/document/11029000", "content": "Recent advances in learning algorithms have identified loss surface sharpness as an effective metric for reducing the generalization gap. Building on this principle, Sharpness -Aware Minimization ( SAM ) was introduced to improve model generalization and has achieved state-of-the-art performance through its weight perturbation and updating steps. However, SAM's perturbation is based solely on ..."}
+{"idx": 7, "title": "Sharpness-Aware Minimization Alone can Improve Adversarial Robustness", "date": "", "ddg_snippet": "Sharpness -Aware Minimization ( SAM ) (Foret et al., 2020) is a novel training framework that improves model generalization by simultaneously minimizing loss value and loss sharpness .", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2305.05392", "content": "Sharpness -Aware Minimization ( SAM ) (Foret et al., 2020) is a novel training framework that improves model generalization by simultaneously minimizing loss value and loss sharpness ."}
+{"idx": 8, "title": "GitHub - Jannoshh/simple-sam: Sharpness-Aware Minimization for ...", "date": "", "ddg_snippet": "Shortened abstract : Optimizing only the training loss value , as is commonly done, can easily lead to suboptimal model quality. Motivated by the connection between geometry of the loss landscape and generalization, SAM is a novel, effective procedure for instead simultaneously minimizing loss value and loss sharpness .", "subpage_snippet": "", "source": "github.com", "link": "https://github.com/Jannoshh/simple-sam", "content": "Shortened abstract : Optimizing only the training loss value , as is commonly done, can easily lead to suboptimal model quality. Motivated by the connection between geometry of the loss landscape and generalization, SAM is a novel, effective procedure for instead simultaneously minimizing loss value and loss sharpness ."}
+{"idx": 9, "title": "S -A M E IMPROVING GENERALIZATION - OpenReview", "date": "", "ddg_snippet": "We introduce Sharpness -Aware Minimization ( SAM ), a novel procedure that improves model generalization by simultaneously minimizing loss value and loss sharpness . SAM functions by seeking parameters that lie in neighborhoods having uniformly low loss value (rather than parameters that only themselves have low loss value , as illustrated in the middle and righthand images of Figure 1), and can be ...", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/references/pdf?id=n3t3jviyLg", "content": "We introduce Sharpness -Aware Minimization ( SAM ), a novel procedure that improves model generalization by simultaneously minimizing loss value and loss sharpness . SAM functions by seeking parameters that lie in neighborhoods having uniformly low loss value (rather than parameters that only themselves have low loss value , as illustrated in the middle and righthand images of Figure 1), and can be ..."}
diff --git a/data/sampled_jsons/sitearxiv.org_Open_the_Black_Box_Step-based_Policy_Updates_for_Temporally-Correlated_Episodic_Reinfo.jsonl b/data/sampled_jsons/sitearxiv.org_Open_the_Black_Box_Step-based_Policy_Updates_for_Temporally-Correlated_Episodic_Reinfo.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..46a2962c7a57f1ffbe33c853a509b2cbe9a86661
--- /dev/null
+++ b/data/sampled_jsons/sitearxiv.org_Open_the_Black_Box_Step-based_Policy_Updates_for_Temporally-Correlated_Episodic_Reinfo.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Open the Black Box: Step-based Policy Updates for Temporally ... arXiv:2401.11437v1 [cs.LG] 21 Jan 2024 Chunking the Critic: A Transformer-based Soft Actor-Critic ... A arXiv:2410.09536v4 [cs.LG] 15 Mar 2025 Overcoming Slow Decision Frequencies in Continuous Control ... A arXiv:2502.07005v3 [cs.LG] 13 Feb 2025 BMP: Bridging the Gap between B-Spline and Movement Primitives", "date": "", "ddg_snippet": "Jan 21, 2024 · Abstract page for arXiv paper 2401.11437: Open the Black Box : Step - based Policy Updates for Temporally - Correlated Episodic Reinforcement Learning ABSTRACT Current advancements in reinforcement learning (RL) have predominantly fo-cused on learning step-based policies that generate actions for each perceived state. While these methods eficiently leverage step information from environ-mental interaction, they often ignore the temporal correlation between actions, resulting in ineficient exploration and unsmooth trajectories that are ... Abstract Soft Actor-Critic (SAC) critically depends on its critic network, which typically evaluates a single state-action pair to guide policy updates. Using N-step returns is a common practice to reduce the bias in the target values of the critic. ABSTRACT This work introduces Transformer-based Off-Policy Episodic Reinforcement Learning (TOP-ERL), a novel algorithm that enables off-policy updates in an ERL framework. In ERL, policies predict entire action trajectories over multiple time steps instead of single per-step actions. These trajectories are typically parame-terized by trajectory generators such as Movement Primitives (MP ... Abstract Reinforcement learning (RL) is rapidly reaching and surpassing human-level control capabilities. However, state-of-the-art RL algorithms often require timesteps and reaction times significantly faster than human capabilities, which is impractical in real-world settings and typically necessitates specialized hardware. We introduce Sequence Reinforcement Learning (SRL), an RL algorithm ... ABSTRACT Manipulating objects with varying geometries and deformable objects is a major challenge in robotics. Tasks such as insertion with different objects or cloth hang-ing require precise control and effective modelling of complex dynamics. In this work, we frame this problem through the lens of a heterogeneous graph that com-prises smaller sub-graphs, such as actuators and objects ... Nov 15, 2024 · Abstract This work introduces B-spline Movement Primitives (BMPs), a new Movement Primitive (MP) variant that leverages B-splines for motion representation. B-splines are a well-known concept in motion planning due to their ability to generate complex, smooth trajectories with only a few control points while satisfying boundary conditions, i.e., passing through a specified desired position ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2401.11437", "content": "Jan 21, 2024 · Abstract page for arXiv paper 2401.11437: Open the Black Box : Step - based Policy Updates for Temporally - Correlated Episodic Reinforcement Learning ABSTRACT Current advancements in reinforcement learning (RL) have predominantly fo-cused on learning step-based policies that generate actions for each perceived state. While these methods eficiently leverage step information from environ-mental interaction, they often ignore the temporal correlation between actions, resulting in ineficient exploration and unsmooth trajectories that are ... Abstract Soft Actor-Critic (SAC) critically depends on its critic network, which typically evaluates a single state-action pair to guide policy updates. Using N-step returns is a common practice to reduce the bias in the target values of the critic. ABSTRACT This work introduces Transformer-based Off-Policy Episodic Reinforcement Learning (TOP-ERL), a novel algorithm that enables off-policy updates in an ERL framework. In ERL, policies predict entire action trajectories over multiple time steps instead of single per-step actions. These trajectories are typically parame-terized by trajectory generators such as Movement Primitives (MP ... Abstract Reinforcement learning (RL) is rapidly reaching and surpassing human-level control capabilities. However, state-of-the-art RL algorithms often require timesteps and reaction times significantly faster than human capabilities, which is impractical in real-world settings and typically necessitates specialized hardware. We introduce Sequence Reinforcement Learning (SRL), an RL algorithm ... ABSTRACT Manipulating objects with varying geometries and deformable objects is a major challenge in robotics. Tasks such as insertion with different objects or cloth hang-ing require precise control and effective modelling of complex dynamics. In this work, we frame this problem through the lens of a heterogeneous graph that com-prises smaller sub-graphs, such as actuators and objects ... Nov 15, 2024 · Abstract This work introduces B-spline Movement Primitives (BMPs), a new Movement Primitive (MP) variant that leverages B-splines for motion representation. B-splines are a well-known concept in motion planning due to their ability to generate complex, smooth trajectories with only a few control points while satisfying boundary conditions, i.e., passing through a specified desired position ..."}
+{"idx": 1, "title": "U pdates for t emporally -c orrelated e pisodic", "date": "", "ddg_snippet": "Open the black box : step - based policy updates for temporally - correlated episodic .Current advancements in reinforcement learning (RL) have predominantly fo-cused on learning step - based policies that generate actions for each perceived state.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2401.11437", "content": "Open the black box : step - based policy updates for temporally - correlated episodic .Current advancements in reinforcement learning (RL) have predominantly fo-cused on learning step - based policies that generate actions for each perceived state."}
+{"idx": 2, "title": "MoRe-ERL: Learning Motion Residuals using Episodic ...", "date": "", "ddg_snippet": "[25] G. Li, H. Zhou, D. Roth, S. Thilges, F. Otto, R. Lioutikov, and G. Neumann, “ Open the black box : Step - based policy updates for temporally - correlated episodic reinforcement learning ,” in The Twelfth International Conference on Learning Representations, 2024.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2508.01409", "content": "[25] G. Li, H. Zhou, D. Roth, S. Thilges, F. Otto, R. Lioutikov, and G. Neumann, “ Open the black box : Step - based policy updates for temporally - correlated episodic reinforcement learning ,” in The Twelfth International Conference on Learning Representations, 2024."}
+{"idx": 3, "title": "arXiv:2401.11437v1 [cs.LG] 21 Jan 2024", "date": "", "ddg_snippet": "ABSTRACT Current advancements in reinforcement learning (RL) have predominantly fo-cused on learning step-based policies that generate actions for each perceived state. While these methods eficiently leverage step information from environ-mental interaction, they often ignore the temporal correlation between actions, resulting in ineficient exploration and unsmooth trajectories that are ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2401.11437.pdf", "content": "ABSTRACT Current advancements in reinforcement learning (RL) have predominantly fo-cused on learning step-based policies that generate actions for each perceived state. While these methods eficiently leverage step information from environ-mental interaction, they often ignore the temporal correlation between actions, resulting in ineficient exploration and unsmooth trajectories that are ..."}
+{"idx": 4, "title": "Chunking the Critic: A Transformer-based Soft Actor-Critic ...", "date": "", "ddg_snippet": "Abstract Soft Actor-Critic (SAC) critically depends on its critic network, which typically evaluates a single state-action pair to guide policy updates. Using N-step returns is a common practice to reduce the bias in the target values of the critic.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2503.03660v1", "content": "Abstract Soft Actor-Critic (SAC) critically depends on its critic network, which typically evaluates a single state-action pair to guide policy updates. Using N-step returns is a common practice to reduce the bias in the target values of the critic."}
+{"idx": 5, "title": "A arXiv:2410.09536v4 [cs.LG] 15 Mar 2025", "date": "", "ddg_snippet": "ABSTRACT This work introduces Transformer-based Off-Policy Episodic Reinforcement Learning (TOP-ERL), a novel algorithm that enables off-policy updates in an ERL framework. In ERL, policies predict entire action trajectories over multiple time steps instead of single per-step actions. These trajectories are typically parame-terized by trajectory generators such as Movement Primitives (MP ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2410.09536", "content": "ABSTRACT This work introduces Transformer-based Off-Policy Episodic Reinforcement Learning (TOP-ERL), a novel algorithm that enables off-policy updates in an ERL framework. In ERL, policies predict entire action trajectories over multiple time steps instead of single per-step actions. These trajectories are typically parame-terized by trajectory generators such as Movement Primitives (MP ..."}
+{"idx": 6, "title": "A arXiv:2502.07005v3 [cs.LG] 13 Feb 2025", "date": "", "ddg_snippet": "ABSTRACT Manipulating objects with varying geometries and deformable objects is a major challenge in robotics. Tasks such as insertion with different objects or cloth hang-ing require precise control and effective modelling of complex dynamics. In this work, we frame this problem through the lens of a heterogeneous graph that com-prises smaller sub-graphs, such as actuators and objects ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/pdf/2502.07005v3", "content": "ABSTRACT Manipulating objects with varying geometries and deformable objects is a major challenge in robotics. Tasks such as insertion with different objects or cloth hang-ing require precise control and effective modelling of complex dynamics. In this work, we frame this problem through the lens of a heterogeneous graph that com-prises smaller sub-graphs, such as actuators and objects ..."}
+{"idx": 7, "title": "BMP: Bridging the Gap between B-Spline and Movement Primitives", "date": "", "ddg_snippet": "Nov 15, 2024 · Abstract This work introduces B-spline Movement Primitives (BMPs), a new Movement Primitive (MP) variant that leverages B-splines for motion representation. B-splines are a well-known concept in motion planning due to their ability to generate complex, smooth trajectories with only a few control points while satisfying boundary conditions, i.e., passing through a specified desired position ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2411.10336v1", "content": "Nov 15, 2024 · Abstract This work introduces B-spline Movement Primitives (BMPs), a new Movement Primitive (MP) variant that leverages B-splines for motion representation. B-splines are a well-known concept in motion planning due to their ability to generate complex, smooth trajectories with only a few control points while satisfying boundary conditions, i.e., passing through a specified desired position ..."}
+{"idx": 8, "title": "Overcoming Slow Decision Frequencies in Continuous Control ...", "date": "", "ddg_snippet": "Abstract Reinforcement learning (RL) is rapidly reaching and surpassing human-level control capabilities. However, state-of-the-art RL algorithms often require timesteps and reaction times significantly faster than human capabilities, which is impractical in real-world settings and typically necessitates specialized hardware. We introduce Sequence Reinforcement Learning (SRL), an RL algorithm ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2410.08979v5", "content": "Abstract Reinforcement learning (RL) is rapidly reaching and surpassing human-level control capabilities. However, state-of-the-art RL algorithms often require timesteps and reaction times significantly faster than human capabilities, which is impractical in real-world settings and typically necessitates specialized hardware. We introduce Sequence Reinforcement Learning (SRL), an RL algorithm ..."}
+{"idx": 9, "title": "Overcoming Slow Decision Frequencies in Continuous Control...", "date": "", "ddg_snippet": "Open the black box : Step - based policy updates for temporally - correlated episodic reinforcement learning .", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2410.08979v3", "content": "Open the black box : Step - based policy updates for temporally - correlated episodic reinforcement learning ."}
diff --git a/data/sampled_jsons/sitearxiv.orghtml2412.06329_Equation_(6).jsonl b/data/sampled_jsons/sitearxiv.orghtml2412.06329_Equation_(6).jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..44296beb3852e677fd5005ac7c820508378fa88f
--- /dev/null
+++ b/data/sampled_jsons/sitearxiv.orghtml2412.06329_Equation_(6).jsonl
@@ -0,0 +1 @@
+{"idx": 0, "title": "", "date": "", "ddg_snippet": "", "subpage_snippet": "", "source": "", "link": "", "content": ""}
diff --git a/data/sampled_jsons/systematization_operationalization_Evaluating_Generative_AI_Systems_Wallach.jsonl b/data/sampled_jsons/systematization_operationalization_Evaluating_Generative_AI_Systems_Wallach.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..8c8cdac10e343a47f066324d8f4948de1f619955
--- /dev/null
+++ b/data/sampled_jsons/systematization_operationalization_Evaluating_Generative_AI_Systems_Wallach.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Evaluating Generative AI Systems is a Social Science", "date": "", "ddg_snippet": "... academia, industry, and government, there is an increasing awareness that the measurement tasks involved in evaluating generative AI (GenAI) systems ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2411.10939v1", "content": "... academia, industry, and government, there is an increasing awareness that the measurement tasks involved in evaluating generative AI (GenAI) systems ..."}
+{"idx": 1, "title": "Position: Evaluating Generative AI Systems is a Social Science", "date": "", "ddg_snippet": "The measurement tasks involved in evaluating generative AI (GenAI) systems are especially difficult, leading to what has been described as “a ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2502.00561v1", "content": "The measurement tasks involved in evaluating generative AI (GenAI) systems are especially difficult, leading to what has been described as “a ..."}
+{"idx": 2, "title": "Multi-Agent LLMs as Ethics Advocates for AI based Systems", "date": "", "ddg_snippet": "Yamani et al., [ 9 ] used three state-of-the-art LLMs to generate 3000 user stories for 100 AI -based systems , constructing the UStAI dataset.", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2507.08392v3", "content": "Yamani et al., [ 9 ] used three state-of-the-art LLMs to generate 3000 user stories for 100 AI -based systems , constructing the UStAI dataset."}
+{"idx": 3, "title": "A Shared Standard for Valid Measurement of Generative AI", "date": "", "ddg_snippet": "The valid measurement of generative AI (GenAI) systems ’ capabilities, risks, and impacts forms the bedrock of our ability to evaluate these ...", "subpage_snippet": "", "source": "www.microsoft.com", "link": "https://www.microsoft.com/en-us/research/publication/a-shared-standard-for-valid-measurement-of-generative-ai-systems-capabilities-risks-and-impacts/", "content": "The valid measurement of generative AI (GenAI) systems ’ capabilities, risks, and impacts forms the bedrock of our ability to evaluate these ..."}
+{"idx": 4, "title": "The Impossibility of Fair LLMs", "date": "", "ddg_snippet": "... systems and concerns about their negative societal impacts, researchers have developed compelling, nuanced technical frameworks to formalize ethical ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2406.03198v2", "content": "... systems and concerns about their negative societal impacts, researchers have developed compelling, nuanced technical frameworks to formalize ethical ..."}
+{"idx": 5, "title": "Towards Interactive Evaluations for Interaction Harms in", "date": "", "ddg_snippet": "We begin by examining contemporary approaches to ethics and safety evaluations of generative AI systems — their methodologies, primary focus ...", "subpage_snippet": "", "source": "knightcolumbia.org", "link": "https://knightcolumbia.org/content/towards-interactive-evaluations-for-interaction-harms-in-human-ai-systems", "content": "We begin by examining contemporary approaches to ethics and safety evaluations of generative AI systems — their methodologies, primary focus ..."}
+{"idx": 6, "title": "Rishi Bommasani's research works | Stanford University and", "date": "", "ddg_snippet": "... generative AI creates unique challenges for system safety engineering and measurement science, the field can draw valuable insights from the ...", "subpage_snippet": "", "source": "www.researchgate.net", "link": "https://www.researchgate.net/scientific-contributions/Rishi-Bommasani-2162493649", "content": "... generative AI creates unique challenges for system safety engineering and measurement science, the field can draw valuable insights from the ..."}
+{"idx": 7, "title": "AI Frontiers: Rethinking intelligence with Ashley Llorens and", "date": "", "ddg_snippet": "... of people and rats to create prompts for evaluating large language models; and the case for the development of a “prefrontal cortex” for AI ...", "subpage_snippet": "", "source": "www.microsoft.com", "link": "https://www.microsoft.com/en-us/research/podcast/ai-frontiers-rethinking-intelligence-with-ashley-llorens-and-ida-momennejad/", "content": "... of people and rats to create prompts for evaluating large language models; and the case for the development of a “prefrontal cortex” for AI ..."}
+{"idx": 8, "title": "HumanAgencyBench: Scalable Evaluation of Human Agency Support", "date": "", "ddg_snippet": "... evaluation methodologies allow us to study it systematically by using LLMs to simulate user queries (i.e., tests), to validate the quality of those ...", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/html/2509.08494v1", "content": "... evaluation methodologies allow us to study it systematically by using LLMs to simulate user queries (i.e., tests), to validate the quality of those ..."}
+{"idx": 9, "title": "The Ethics of AI Ethics: An Evaluation of Guidelines | Minds", "date": "", "ddg_snippet": "Designed as a semi-systematic evaluation, this paper analyzes and compares 22 guidelines, highlighting overlaps but also omissions.", "subpage_snippet": "", "source": "link.springer.com", "link": "https://link.springer.com/article/10.1007/s11023-020-09517-8", "content": "Designed as a semi-systematic evaluation, this paper analyzes and compares 22 guidelines, highlighting overlaps but also omissions."}
diff --git a/data/sampled_jsons/vh9yEPLeyD_Can_We_Leave_Deepfake_Data_Behind_hyperparameters_beta_gamma_section_4.1.jsonl b/data/sampled_jsons/vh9yEPLeyD_Can_We_Leave_Deepfake_Data_Behind_hyperparameters_beta_gamma_section_4.1.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..cca292699db67af5bf61aa37e1151c864840259c
--- /dev/null
+++ b/data/sampled_jsons/vh9yEPLeyD_Can_We_Leave_Deepfake_Data_Behind_hyperparameters_beta_gamma_section_4.1.jsonl
@@ -0,0 +1,10 @@
+{"idx": 0, "title": "Hyperparameter (machine learning) - Wikipedia", "date": "", "ddg_snippet": "In machine learning, a hyperparameter is a parameter that can be set in order to define any configurable part of a model's learning process. Hyperparameters can be classified as either model hyperparameters or algorithm hyperparameters .", "subpage_snippet": "", "source": "en.wikipedia.org", "link": "https://en.wikipedia.org/wiki/Hyperparameter_(machine_learning)", "content": "In machine learning, a hyperparameter is a parameter that can be set in order to define any configurable part of a model's learning process. Hyperparameters can be classified as either model hyperparameters or algorithm hyperparameters ."}
+{"idx": 1, "title": "Can We Leave Deepfake Data Behind in Training", "date": "", "ddg_snippet": "EfficientB4 can be treated as the baseline of only using deepfake data . X-ray (CBI) and SBI are the baselines of only using one specific type of blendfake data .", "subpage_snippet": "", "source": "openreview.net", "link": "https://openreview.net/pdf?id=vh9yEPLeyD", "content": "EfficientB4 can be treated as the baseline of only using deepfake data . X-ray (CBI) and SBI are the baselines of only using one specific type of blendfake data ."}
+{"idx": 2, "title": "Can We Leave Deepfake Data Behind in Training Deepfake Detector?", "date": "", "ddg_snippet": "vh 9 yEPLeyD @OpenReview. Total: 1.Therefore, a critical question arises: Can we leave deepfake behind and rely solely on blendfake data to train an effective deepfake detector?", "subpage_snippet": "", "source": "papers.cool", "link": "https://papers.cool/venue/vh9yEPLeyD@OpenReview", "content": "vh 9 yEPLeyD @OpenReview. Total: 1.Therefore, a critical question arises: Can we leave deepfake behind and rely solely on blendfake data to train an effective deepfake detector?"}
+{"idx": 3, "title": "Can We Leave Deepfake Data Behind in Training Deepfake Detector?", "date": "", "ddg_snippet": "Therefore, a critical question arises: Can we leave deepfake behind and rely solely on blendfake data to train an effective deepfake detector?", "subpage_snippet": "", "source": "arxiv.org", "link": "https://arxiv.org/abs/2408.17052", "content": "Therefore, a critical question arises: Can we leave deepfake behind and rely solely on blendfake data to train an effective deepfake detector?"}
+{"idx": 4, "title": "DF40: Toward Next-Generation Deepfake Detection (2024)", "date": "", "ddg_snippet": "Can We Leave Deepfake Data Behind in Training Deepfake Detector?1.4K. Podcast. •Proceedings Article•DOI.", "subpage_snippet": "", "source": "scispace.com", "link": "https://scispace.com/papers/df40-toward-next-generation-deepfake-detection-273mzyvz9x", "content": "Can We Leave Deepfake Data Behind in Training Deepfake Detector?1.4K. Podcast. •Proceedings Article•DOI."}
+{"idx": 5, "title": "A Guide on XGBoost hyperparameters tuning", "date": "", "ddg_snippet": "Explore and run machine learning code with Kaggle Notebooks | Using data from Wholesale customers Data Set.", "subpage_snippet": "", "source": "www.kaggle.com", "link": "https://www.kaggle.com/code/prashant111/a-guide-on-xgboost-hyperparameters-tuning", "content": "Explore and run machine learning code with Kaggle Notebooks | Using data from Wholesale customers Data Set."}
+{"idx": 6, "title": "Can We Leave Deepfake Data Behind in Training Deepfake Detector?", "date": "", "ddg_snippet": "vh 9 yEPLeyD .This raises a critical question: can deepfake data be entirely discarded? This paper investigates this question and finds this to be counter-intuitive.", "subpage_snippet": "", "source": "deep-diver.github.io", "link": "https://deep-diver.github.io/neurips2024/posters/vh9yepleyd/", "content": "vh 9 yEPLeyD .This raises a critical question: can deepfake data be entirely discarded? This paper investigates this question and finds this to be counter-intuitive."}
+{"idx": 7, "title": "Difference Between Model Parameters VS HyperParameters", "date": "", "ddg_snippet": "The two most confusing terms in Machine Learning are Model Parameters and Hyperparameters . In this post, we will try to understand what these terms mean and how they are different from each other.", "subpage_snippet": "", "source": "www.geeksforgeeks.org", "link": "https://www.geeksforgeeks.org/machine-learning/difference-between-model-parameters-vs-hyperparameters/", "content": "The two most confusing terms in Machine Learning are Model Parameters and Hyperparameters . In this post, we will try to understand what these terms mean and how they are different from each other."}
+{"idx": 8, "title": "towardsdatascience.com/parameters-and- hyperparameters -aa609601...", "date": "", "ddg_snippet": "Behind the training scene, parameters are continuously being updated and...", "subpage_snippet": "", "source": "towardsdatascience.com", "link": "https://towardsdatascience.com/parameters-and-hyperparameters-aa609601a9ac/", "content": "Behind the training scene, parameters are continuously being updated and..."}
+{"idx": 9, "title": "ГДЗ Английский язык 7 класс Биболетова, Трубанева, 2019 на...", "date": "", "ddg_snippet": "SECTION 9. How can we communicate with each other. SECTION 4 . Why do people like to compete.", "subpage_snippet": "", "source": "Reshalka.com", "link": "https://Reshalka.com/uchebniki/7-klass/english/biboletova1", "content": "SECTION 9. How can we communicate with each other. SECTION 4 . Why do people like to compete."}