title
stringlengths
15
138
paper_url
stringlengths
42
42
authors
listlengths
1
35
type
stringclasses
3 values
primary_area
stringclasses
20 values
abstract
large_stringlengths
480
3.09k
keywords
listlengths
1
27
TL;DR
large_stringlengths
21
250
submission_number
int64
5
9.5k
arxiv_id
stringlengths
10
10
arxiv_id_source
stringclasses
2 values
embedding
listlengths
768
768
TabR: Tabular Deep Learning Meets Nearest Neighbors
https://openreview.net/forum?id=rhgIgTSSxW
[ "Yury Gorishniy", "Ivan Rubachev", "Nikolay Kartashev", "Daniil Shlenskii", "Akim Kotelnikov", "Artem Babenko" ]
Poster
general machine learning (i.e., none of the above)
Deep learning (DL) models for tabular data problems (e.g. classification, regression) are currently receiving increasingly more attention from researchers. However, despite the recent efforts, the non-DL algorithms based on gradient-boosted decision trees (GBDT) remain a strong go-to solution for these problems. One of...
[ "tabular", "tabular data", "architecture", "deep learning", "neural networks" ]
TabR is a new tabular DL model with a k-nearest-neighbors-like component and strong results on public benchmarks.
9,502
2307.14338
title_judge
[ -0.05746728554368019, -0.024489738047122955, -0.017151275649666786, 0.06507784128189087, 0.028546025976538658, -0.008366950787603855, -0.004765615798532963, 0.0023200043942779303, -0.02343326061964035, -0.04401092231273651, -0.02110673487186432, -0.012399819679558277, -0.07017732411623001, ...
SaNN: Simple Yet Powerful Simplicial-aware Neural Networks
https://openreview.net/forum?id=eUgS9Ig8JG
[ "Sravanthi Gurugubelli", "Sundeep Prabhakar Chepuri" ]
Spotlight
learning on graphs and other geometries & topologies
Simplicial neural networks (SNNs) are deep models for higher-order graph representation learning. SNNs learn low-dimensional embeddings of simplices in a simplicial complex by aggregating features of their respective upper, lower, boundary, and coboundary adjacent simplices. The aggregation in SNNs is carried out durin...
[ "Graph Neural Networks", "Higher-order Representation Learning", "Simplicial Complexes", "Simplicial Neural Networks", "Weisfeiler-Lehman Isomorphism Test" ]
null
9,491
null
null
[ -0.029635116457939148, -0.02202700451016426, 0.014880966395139694, 0.04183591157197952, -0.009340484626591206, 0.026812955737113953, 0.013645652681589127, 0.02132989466190338, -0.02486003376543522, -0.0353904664516449, 0.015543965622782707, -0.012865939177572727, -0.07329703122377396, 0.01...
Boosting of Thoughts: Trial-and-Error Problem Solving with Large Language Models
https://openreview.net/forum?id=qBL04XXex6
[ "Sijia Chen", "Baochun Li", "Di Niu" ]
Poster
generative models
The reasoning performance of Large Language Models (LLMs) on a wide range of problems critically relies on chain-of-thought prompting, which involves providing a few chain of thought demonstrations as exemplars in prompts. Recent work, e.g., Tree of Thoughts, has pointed out the importance of exploration and self-evalu...
[ "Large Language Models; Prompt Engineering; Boosting Mechanism;" ]
null
9,482
2402.11140
title_snapshot
[ -0.020473770797252655, -0.016034100204706192, -0.022892706096172333, 0.04547521471977234, 0.04395535960793495, 0.021775787696242332, 0.030156390741467476, 0.01553252711892128, -0.034200992435216904, -0.0022243394050747156, -0.03878597915172577, 0.03016478754580021, -0.0658145546913147, -0....
Beyond Memorization: Violating Privacy via Inference with Large Language Models
https://openreview.net/forum?id=kmn0BhQk7p
[ "Robin Staab", "Mark Vero", "Mislav Balunovic", "Martin Vechev" ]
Spotlight
societal considerations including fairness, safety, privacy
Current privacy research on large language models (LLMs) primarily focuses on the issue of extracting memorized training data. At the same time, models’ inference capabilities have increased drastically. This raises the key question of whether current LLMs could violate individuals’ privacy by inferring personal attrib...
[ "Privacy", "Large Language Models" ]
We present the first comprehensive study on the capabilities of pretrained LLMs to infer personal attributes from texts given at inference.
9,451
2310.07298
title_snapshot
[ 0.0018264753744006157, -0.000775889668148011, -0.00885173212736845, 0.04171423614025116, 0.04986639320850372, -0.011038335971534252, 0.03983701765537262, 0.01709677465260029, -0.017802106216549873, 0.012574026361107826, -0.02018856443464756, 0.035563718527555466, -0.051449358463287354, -0....
Locality Sensitive Sparse Encoding for Learning World Models Online
https://openreview.net/forum?id=i8PjQT3Uig
[ "Zichen Liu", "Chao Du", "Wee Sun Lee", "Min Lin" ]
Poster
reinforcement learning
Acquiring an accurate world model $\textit{online}$ for model-based reinforcement learning (MBRL) is challenging due to data nonstationarity, which typically causes catastrophic forgetting for neural networks (NNs). From the online learning perspective, a Follow-The-Leader (FTL) world model is desirable, which optimall...
[ "model-based rl", "online learning", "incremental learning", "catastrophic forgetting" ]
null
9,441
2401.13034
title_snapshot
[ -0.04537220671772957, -0.010137720964848995, 0.017094457522034645, 0.04123565927147865, 0.025048932060599327, 0.04178636893630028, 0.00010264802403980866, 0.026731934398412704, -0.04688441753387451, -0.024245265871286392, -0.017856601625680923, -0.01075384858995676, -0.07858645915985107, 0...
Enhancing Neural Subset Selection: Integrating Background Information into Set Representations
https://openreview.net/forum?id=eepoE7iLpL
[ "Binghui Xie", "Yatao Bian", "Kaiwen Zhou", "Yongqiang Chen", "Peilin Zhao", "Bo Han", "Wei Meng", "James Cheng" ]
Poster
representation learning for computer vision, audio, language, and other modalities
Learning neural subset selection tasks, such as compound selection in AI-aided drug discovery, have become increasingly pivotal across diverse applications. The existing methodologies in the field primarily concentrate on constructing models that capture the relationship between utility function values and subsets with...
[ "Neural Set Function", "Hierarchical Structure", "Invariance", "Subset Selection" ]
null
9,406
2402.03139
title_snapshot
[ -0.019677523523569107, -0.0244857557117939, -0.02345975488424301, 0.04267389327287674, 0.04543006420135498, 0.01501544751226902, -0.0016627429286018014, -0.029224881902337074, -0.030251363292336464, -0.04217834025621414, -0.03118780441582203, 0.0462665855884552, -0.0740964338183403, -0.009...
Bridging Vision and Language Spaces with Assignment Prediction
https://openreview.net/forum?id=lK2V2E2MNv
[ "Jungin Park", "Jiyoung Lee", "Kwanghoon Sohn" ]
Poster
representation learning for computer vision, audio, language, and other modalities
This paper introduces VLAP, a novel approach that bridges pretrained vision models and large language models (LLMs) to make frozen LLMs understand the visual world. VLAP transforms the embedding space of pretrained vision models into the LLMs' word embedding space using a single linear layer for efficient and general-p...
[ "Multimodal learning", "vision-language tasks", "frozen LLMs", "optimal transport", "assignment prediction" ]
This paper presents to bridge frozen image encoders and large language models (LLMs) for grounding LLMs to images.
9,396
2404.09632
title_snapshot
[ 0.013795919716358185, 0.018235132098197937, 0.00701660942286253, 0.048582643270492554, 0.03553276136517525, 0.039255619049072266, 0.023923786357045174, 0.026098094880580902, -0.03392104431986809, -0.013619310222566128, -0.04442369192838669, 0.014564957469701767, -0.0775119960308075, -0.009...
Generative Judge for Evaluating Alignment
https://openreview.net/forum?id=gtkFw6sZGS
[ "Junlong Li", "Shichao Sun", "Weizhe Yuan", "Run-Ze Fan", "hai zhao", "Pengfei Liu" ]
Poster
generative models
The rapid development of Large Language Models (LLMs) has substantially expanded the range of tasks they can address. In the field of Natural Language Processing (NLP), researchers have shifted their focus from conventional NLP tasks (e.g., sequence tagging and parsing) towards tasks that revolve around aligning with h...
[ "Generative", "Evaluation", "Alignment" ]
We release Auto-J, a cutting-edge, flexible and interpretable judge with 13B parameters, to evaluate alignment in various real-world scenarios.
9,392
2310.05470
title_snapshot
[ 0.00048703342326916754, -0.04823862761259079, -0.03456166014075279, 0.036054737865924835, 0.018531499430537224, 0.028080958873033524, 0.032217152416706085, 0.028387105092406273, -0.00561843067407608, -0.018411174416542053, -0.02464251220226288, 0.03848639875650406, -0.08764978498220444, -0...
Rethinking and Extending the Probabilistic Inference Capacity of GNNs
https://openreview.net/forum?id=7vVWiCrFnd
[ "Tuo Xu", "Lei Zou" ]
Poster
learning on graphs and other geometries & topologies
Designing expressive Graph Neural Networks (GNNs) is an important topic in graph machine learning fields. Despite the existence of numerous approaches proposed to enhance GNNs based on Weisfeiler-Lehman (WL) tests, what GNNs can and cannot learn still lacks a deeper understanding. This paper adopts a fundamentally diff...
[ "graph neural networks", "expressiveness", "approximate inference" ]
Discuss and extend GNNs' expressive power for probabilistic inference.
9,389
null
null
[ -0.023964818567037582, -0.035738229751586914, 0.008395818062126637, 0.050926826894283295, 0.03230826184153557, 0.013713599182665348, 0.02981681562960148, 0.017039697617292404, -0.006786307319998741, -0.022156130522489548, 0.019732076674699783, -0.026444463059306145, -0.06755448132753372, -...
Learning model uncertainty as variance-minimizing instance weights
https://openreview.net/forum?id=bDWXhzZT40
[ "Nishant Jain", "Karthikeyan Shanmugam", "Pradeep Shenoy" ]
Poster
general machine learning (i.e., none of the above)
Predictive uncertainty--a model’s self-awareness regarding its accuracy on an input--is key for both building robust models via training interventions and for test-time applications such as selective classification. We propose a novel instance-conditional reweighting approach that captures predictive uncertainty using ...
[ "loss reweighting", "epistemic uncertainty", "bi-level optimization", "model calibration", "bayesian neural networks" ]
null
9,383
null
null
[ -0.0009647338301874697, 0.005153102800250053, 0.005902040749788284, 0.03594478592276573, 0.022968364879488945, 0.06027489900588989, 0.0353173166513443, -0.015521418303251266, -0.04422784596681595, -0.04910435900092125, -0.03709302470088005, 0.014822098426520824, -0.0922713652253151, -0.002...
Controlled Text Generation via Language Model Arithmetic
https://openreview.net/forum?id=SLw9fp4yI6
[ "Jasper Dekoninck", "Marc Fischer", "Luca Beurer-Kellner", "Martin Vechev" ]
Spotlight
generative models
As Large Language Models (LLMs) are deployed more widely, customization with respect to vocabulary, style, and character becomes more important. In this work, we introduce model arithmetic, a novel inference framework for composing and biasing LLMs without the need for model (re)training or highly specific datasets. In...
[ "Controlled text generation", "LLM", "Natural Language Processing" ]
We provide a principled and intuitive way to combine multiple LLMs and bias them towards and away from attributes.
9,377
2311.14479
title_snapshot
[ -0.01569182425737381, 0.0025058314204216003, -0.020247258245944977, 0.04930587112903595, 0.05791759490966797, 0.022024942561984062, 0.02423136681318283, 0.02766329050064087, -0.01735253445804119, -0.0023734956048429012, -0.006549110636115074, 0.03547418490052223, -0.06665977835655212, 0.00...
Forward Learning with Top-Down Feedback: Empirical and Analytical Characterization
https://openreview.net/forum?id=My7lkRNnL9
[ "Ravi Francesco Srinivasan", "Francesca Mignacco", "Martino Sorbaro", "Maria Refinetti", "Avi Cooper", "Gabriel Kreiman", "Giorgia Dellaferrera" ]
Poster
applications to neuroscience & cognitive science
"Forward-only" algorithms, which train neural networks while avoiding a backward pass, have recently gained attention as a way of solving the biologically unrealistic aspects of backpropagation. Here, we first address compelling challenges related to the "forward-only" rules, which include reducing the performance gap ...
[ "Forward-only learning", "Biologically inspired learning", "Artificial neural networks", "Analytical characterization" ]
We discuss "forward-only" algorithms, provide an analytical characterization and test strategies to improve their performance.
9,371
2302.05440
title_snapshot
[ -0.04671306163072586, -0.02512805163860321, -0.009592657908797264, 0.010542310774326324, 0.047780632972717285, 0.022572308778762817, 0.027969099581241608, 0.0015321957180276513, -0.04325928911566734, -0.0378325954079628, 0.020210029557347298, 0.023445602506399155, -0.07100732624530792, -0....
Towards Cross Domain Generalization of Hamiltonian Representation via Meta Learning
https://openreview.net/forum?id=AZGIwqCyYY
[ "Yeongwoo Song", "Hawoong Jeong" ]
Poster
applications to physical sciences (physics, chemistry, biology, etc.)
Recent advances in deep learning for physics have focused on discovering shared representations of target systems by incorporating physics priors or inductive biases into neural networks. While effective, these methods are limited to the system domain, where the type of system remains consistent and thus cannot ensure ...
[ "hamiltonian dynamics", "cross domain generalization", "learning physics", "meta learning" ]
We explore cross domain generalization across dynamical system of diverse functional form of Hamiltonian.
9,361
2212.01168
title_snapshot
[ -0.020135944709181786, 0.018790384754538536, -0.011046200059354305, 0.057020898908376694, 0.023014800623059273, -0.01148935966193676, 0.029192591086030006, 0.007406073622405529, -0.04101651906967163, -0.02773241326212883, 0.018973931670188904, -0.024667266756296158, -0.07230295985937119, 0...
What Makes Good Data for Alignment? A Comprehensive Study of Automatic Data Selection in Instruction Tuning
https://openreview.net/forum?id=BTKAeLqLMw
[ "Wei Liu", "Weihao Zeng", "Keqing He", "Yong Jiang", "Junxian He" ]
Poster
general machine learning (i.e., none of the above)
Instruction tuning is a standard technique employed to align large language models to end tasks and user preferences after the initial pretraining phase. Recent research indicates the critical role of data engineering in instruction tuning -- when appropriately selected, only limited data is necessary to achieve superi...
[ "data selection", "instruction tuning", "large language models" ]
We perform a comprehensive study to understand the characteristics of data samples that are the most effective for alignment, and propose automatic data selection approaches that lead to data-efficient instruction tuning
9,349
2312.15685
title_snapshot
[ -0.012169824913144112, 0.0018215612508356571, -0.046282265335321426, 0.03236772492527962, 0.03976805508136749, 0.019424254074692726, 0.06754427403211594, 0.030557906255126, 0.006372308824211359, -0.03570816293358803, -0.00793050229549408, 0.043507013469934464, -0.0472395196557045, -0.01316...
Causality-Inspired Spatial-Temporal Explanations for Dynamic Graph Neural Networks
https://openreview.net/forum?id=AJBkfwXh3u
[ "Kesen Zhao", "Liang Zhang" ]
Poster
learning on graphs and other geometries & topologies
Dynamic Graph Neural Networks (DyGNNs) have gained significant popularity in the research of dynamic graphs, but are limited by the low transparency, such that human-understandable insights can hardly be drawn from their predictions. Although a number of existing research have been devoted to investigating the interpre...
[ "Dynamic Graph", "Graph Explanation", "Graph Neural Network", "Causal Inference" ]
To the best of our knowledge, we are the first to explain dynamic graph neural networks.
9,348
null
null
[ 0.012717260979115963, -0.017138080671429634, 0.0037809456698596478, 0.049556631594896317, 0.04741739109158516, 0.04003816843032837, 0.02591804787516594, 0.018012287095189095, -0.013409946113824844, -0.05376554653048515, 0.022360414266586304, -0.002129598753526807, -0.039115726947784424, 0....
Dissecting learning and forgetting in language model finetuning
https://openreview.net/forum?id=tmsqb6WpLz
[ "Xiao Zhang", "Ji Wu" ]
Poster
transfer learning, meta learning, and lifelong learning
Finetuning language models on domain-specific corpus is a common approach to enhance their domain knowledge and capability. While improving performance on domain tasks, it often brings a side-effect of forgetting of the model's general abilities. In this study, we analyze the effects of finetuning on language models by...
[ "language models", "domain adaptation", "catastrophic forgetting" ]
null
9,346
null
null
[ 0.00500105693936348, -0.02625495009124279, -0.021618686616420746, 0.04044949263334274, 0.06151016801595688, -0.01873558945953846, 0.03980281949043274, 0.02726897783577442, -0.03419647365808487, 0.0037499049212783575, -0.021552883088588715, 0.06010643392801285, -0.03854398801922798, 0.01956...
Test-time Adaptation against Multi-modal Reliability Bias
https://openreview.net/forum?id=TPZRq4FALB
[ "Mouxing Yang", "Yunfan Li", "Changqing Zhang", "Peng Hu", "Xi Peng" ]
Poster
unsupervised, self-supervised, semi-supervised, and supervised representation learning
Test-time adaptation (TTA) has emerged as a new paradigm for reconciling distribution shifts across domains without accessing source data. However, existing TTA methods mainly concentrate on uni-modal tasks, overlooking the complexity of multi-modal scenarios. In this paper, we delve into the multi-modal test-time adap...
[ "Test-time adaption", "Imbalanced multi-modal learning" ]
Reveal a new problem named reliability bias for multi-modal TTA, and propose a new method to achieve reliable fusion and robust adaption.
9,339
null
null
[ 0.011494302190840244, -0.023423492908477783, -0.0014195303665474057, 0.07512050122022629, 0.035780150443315506, 0.04741832986474037, 0.0324808768928051, 0.03016664832830429, -0.0023287618532776833, -0.054038263857364655, 0.02175959199666977, 0.03307563066482544, -0.0663301944732666, -0.030...
Mirage: Model-agnostic Graph Distillation for Graph Classification
https://openreview.net/forum?id=78iGZdqxYY
[ "Mridul Gupta", "Sahil Manchanda", "HARIPRASAD KODAMANA", "Sayan Ranu" ]
Poster
learning on graphs and other geometries & topologies
GNNs, like other deep learning models, are data and computation hungry. There is a pressing need to scale training of GNNs on large datasets to enable their usage on low-resource environments. Graph distillation is an effort in that direction with the aim to construct a smaller synthetic training set from the original ...
[ "graph distillation", "graph classification", "frequent pattern mining" ]
An unsupervised and model/hyper-parameter agnostic graph distillation algorithm for graph classification.
9,334
2310.09486
title_snapshot
[ -0.007815618067979813, -0.02467149682343006, -0.020470282062888145, 0.062006350606679916, 0.023842528462409973, 0.005095076747238636, 0.04004882276058197, 0.00790812261402607, -0.031588323414325714, -0.04041801020503044, -0.011901396326720715, -0.0027376434300094843, -0.07972926646471024, ...
On the Learnability of Watermarks for Language Models
https://openreview.net/forum?id=9k0krNzvlV
[ "Chenchen Gu", "Xiang Lisa Li", "Percy Liang", "Tatsunori Hashimoto" ]
Poster
general machine learning (i.e., none of the above)
Watermarking of language model outputs enables statistical detection of model-generated text, which can mitigate harms and misuses of language models. Existing watermarking strategies operate by altering the decoder of an existing language model. In this paper, we ask whether language models can directly learn to gener...
[ "watermarking", "large language models", "distillation" ]
Language models can learn to naturally generate watermarked text, without using any special decoding algorithms.
9,328
2312.04469
title_snapshot
[ -0.00170203170273453, -0.003344718599691987, -0.02847231924533844, 0.07823432981967926, 0.051012903451919556, 0.00314852362498641, 0.032630275934934616, 0.003398573026061058, -0.012820971198379993, -0.002058384707197547, -0.03246670588850975, 0.033901605755090714, -0.04225169122219086, 0.0...
Consistency Training with Learnable Data Augmentation for Graph Anomaly Detection with Limited Supervision
https://openreview.net/forum?id=elMKXvhhQ9
[ "Nan Chen", "Zemin Liu", "Bryan Hooi", "Bingsheng He", "Rizal Fathony", "Jun Hu", "Jia Chen" ]
Spotlight
learning on graphs and other geometries & topologies
Graph Anomaly Detection (GAD) has surfaced as a significant field of research, predominantly due to its substantial influence in production environments. Although existing approaches for node anomaly detection have shown effectiveness, they have yet to fully address two major challenges: operating in settings with limi...
[ "Graph anomaly detection", "consistency training", "learnable data augmentation" ]
null
9,315
null
null
[ 0.014103544875979424, -0.03614644333720207, -0.01578887738287449, 0.07131879031658173, 0.045795049518346786, 0.001075865002349019, 0.04931800067424774, 0.006942900363355875, -0.029316630214452744, -0.04022587090730667, 0.004757594782859087, -0.015829849988222122, -0.1032964363694191, 0.013...
Bellman Optimal Stepsize Straightening of Flow-Matching Models
https://openreview.net/forum?id=Iyve2ycvGZ
[ "Bao Nguyen", "Binh Nguyen", "Viet Anh Nguyen" ]
Poster
generative models
Flow matching is a powerful framework for generating high-quality samples in various applications, especially image synthesis. However, the intensive computational demands of these models, especially during the finetuning process and sampling processes, pose significant challenges for low-resource scenarios. This paper...
[ "flow matching", "generative model", "efficient sampling", "distillation", "responsible ML" ]
This paper introduces Bellman Optimal Step-size Straightening (BOSS) technique for distilling flow-matching generative models while adhering to a computational budget constraint.
9,312
2312.16414
title_snapshot
[ 0.0023783708456903696, -0.01969977840781212, -0.02958757057785988, 0.056870236992836, 0.04038286954164505, 0.06353975832462311, 0.021305253729224205, -0.0136262197047472, -0.041799627244472504, -0.06671983748674393, -0.004898285027593374, -0.05231522396206856, -0.07126540690660477, -0.0131...
Enhancing Tail Performance in Extreme Classifiers by Label Variance Reduction
https://openreview.net/forum?id=6ARlSgun7J
[ "Anirudh Buvanesh", "Rahul Chand", "Jatin Prakash", "Bhawna Paliwal", "Mudit Dhawan", "Neelabh Madan", "Deepesh Hada", "Vidit Jain", "SONU MEHTA", "Yashoteja Prabhu", "Manish Gupta", "Ramachandran Ramjee", "Manik Varma" ]
Poster
unsupervised, self-supervised, semi-supervised, and supervised representation learning
Extreme Classification (XC) architectures, which utilize a massive One-vs-All (OvA) classifier layer at the output, have demonstrated remarkable performance on problems with large label sets. Nonetheless, these architectures falter on tail labels with few representative samples. This phenomenon has been attributed to f...
[ "Extreme Classification", "Extreme Multi-Label Learning" ]
null
9,297
null
null
[ -0.006771616172045469, -0.036368485540151596, -0.0009874040260910988, 0.03520958125591278, 0.04742474853992462, -0.010997247882187366, 0.02056606486439705, -0.01822119951248169, -0.007305160630494356, -0.022930819541215897, -0.03056359849870205, -0.005292092449963093, -0.060889482498168945, ...
Is This the Subspace You Are Looking for? An Interpretability Illusion for Subspace Activation Patching
https://openreview.net/forum?id=Ebt7JgMHv1
[ "Aleksandar Makelov", "Georg Lange", "Atticus Geiger", "Neel Nanda" ]
Poster
visualization or interpretation of learned representations
Mechanistic interpretability aims to attribute high-level model behaviors to specific, interpretable learned features. It is hypothesized that these features manifest as directions or low-dimensional subspaces within activation space. Accordingly, recent studies have explored the identification and manipulation of such...
[ "Mechanistic Interpretability", "Natural Language Processing", "Large Language Models" ]
We show how activation patching can hallucinate meaningful subspaces in a language model by activating dormant pathways.
9,274
2311.17030
title_snapshot
[ -0.0017787369433790445, -0.015876537188887596, -0.0036496773827821016, 0.026157265529036522, 0.057104356586933136, -0.011302768252789974, 0.06107218563556671, -0.008016147650778294, -0.012348785996437073, -0.035270363092422485, -0.02936338074505329, 0.021843325346708298, -0.06432517617940903...
Symmetric Mean-field Langevin Dynamics for Distributional Minimax Problems
https://openreview.net/forum?id=YItWKZci78
[ "Juno Kim", "Kakei Yamamoto", "Kazusato Oko", "Zhuoran Yang", "Taiji Suzuki" ]
Spotlight
optimization
In this paper, we extend mean-field Langevin dynamics to minimax optimization over probability distributions for the first time with symmetric and provably convergent updates. We propose \emph{mean-field Langevin averaged gradient} (MFL-AG), a single-loop algorithm that implements gradient descent ascent in the distrib...
[ "mean-field Langevin dynamics", "minimax optimization", "zero-sum games", "Markov games" ]
null
9,271
2312.01127
title_snapshot
[ -0.022611284628510475, 0.009948745369911194, 0.044614750891923904, 0.02015441097319126, 0.04271610826253891, 0.024143662303686142, 0.012171419337391853, 0.006888951174914837, -0.029769184067845345, -0.06816314905881882, 0.03909756615757942, -0.004533367697149515, -0.0714612528681755, 0.008...
Demonstration-Regularized RL
https://openreview.net/forum?id=lF2aip4Scn
[ "Daniil Tiapkin", "Denis Belomestny", "Daniele Calandriello", "Eric Moulines", "Alexey Naumov", "Pierre Perrault", "Michal Valko", "Pierre Menard" ]
Poster
reinforcement learning
Incorporating expert demonstrations has empirically helped to improve the sample efficiency of reinforcement learning (RL). This paper quantifies theoretically to what extent this extra information reduces RL's sample complexity. In particular, we study the demonstration-regularized reinforcement learning framework tha...
[ "reinforcement learning", "regularization in reinforcement leaning", "learning with demonstrations", "reinforcemenet learning with human feedback" ]
We showed a theoretically efficient way to inject expert demonstrations into RL agent and, moreover, into RLHF.
9,260
2310.17303
title_snapshot
[ -0.02120574563741684, -0.043407633900642395, -0.01618889719247818, 0.06242118030786514, 0.060800906270742416, 0.020388532429933548, 0.020597821101546288, -0.015159987844526768, -0.05165700614452362, -0.028357889503240585, -0.0125327343121171, 0.018507646396756172, -0.08669126033782959, -0....
Multilingual Jailbreak Challenges in Large Language Models
https://openreview.net/forum?id=vESNKdEMGp
[ "Yue Deng", "Wenxuan Zhang", "Sinno Jialin Pan", "Lidong Bing" ]
Poster
societal considerations including fairness, safety, privacy
While large language models (LLMs) exhibit remarkable capabilities across a wide range of tasks, they pose potential safety concerns, such as the ``jailbreak'' problem, wherein malicious instructions can manipulate LLMs to exhibit undesirable behavior. Although several preventive measures have been developed to mitigat...
[ "multilingual", "safety", "large language models" ]
We reveal the presence of multilingual jailbreak challenges within LLMs and propose the Self-Defense framework to mitigate the issue.
9,250
2310.06474
title_snapshot
[ -0.023840511217713356, -0.047415342181921005, -0.029908182099461555, 0.03395756706595421, 0.03871824964880943, 0.019881585612893105, 0.04494781047105789, 0.008515414781868458, -0.04137352481484413, -0.004955085460096598, -0.04960514232516289, 0.019304994493722916, -0.07936786860227585, -0....
Generalized Policy Iteration using Tensor Approximation for Hybrid Control
https://openreview.net/forum?id=csukJcpYDe
[ "Suhan Shetty", "Teng Xue", "Sylvain Calinon" ]
Spotlight
applications to robotics, autonomy, planning
Control of dynamic systems involving hybrid actions is a challenging task in robotics. To address this, we present a novel algorithm called Generalized Policy Iteration using Tensor Train (TTPI) that belongs to the class of Approximate Dynamic Programming (ADP). We use a low-rank tensor approximation technique called ...
[ "Optimal Control", "Hybrid Actions", "Robotics", "Approximate Dynamic Programming", "Tensor Approximation" ]
The paper proposes a novel approximate dynamic programming algorithm that can handle hybrid action space
9,245
null
null
[ -0.05606513097882271, -0.02885339967906475, -0.03463707119226456, 0.042150288820266724, 0.01718045398592949, 0.019540244713425636, 0.013384826481342316, -0.0055174920707941055, -0.024759339168667793, -0.054245155304670334, -0.04266258701682091, 0.02285527065396309, -0.08111023157835007, 0....
$t^3$-Variational Autoencoder: Learning Heavy-tailed Data with Student's t and Power Divergence
https://openreview.net/forum?id=RzNlECeoOB
[ "Juno Kim", "Jaehyuk Kwon", "Mincheol Cho", "Hyunjong Lee", "Joong-Ho Won" ]
Poster
probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)
The variational autoencoder (VAE) typically employs a standard normal prior as a regularizer for the probabilistic latent encoder. However, the Gaussian tail often decays too quickly to effectively accommodate the encoded points, failing to preserve crucial structures hidden in the data. In this paper, we explore the u...
[ "Variational autoencoder", "Information geometry", "Heavy-tail learning", "Generative model" ]
null
9,244
2312.01133
title_snapshot
[ 0.03917887061834335, -0.014440701343119144, 0.017736991867423058, 0.030609918758273125, 0.03900834918022156, 0.02392699010670185, 0.04695601016283035, 0.002285501454025507, -0.015307096764445305, -0.031546931713819504, -0.020012641325592995, -0.01928059197962284, -0.05360564962029457, 0.03...
Rethinking the Power of Graph Canonization in Graph Representation Learning with Stability
https://openreview.net/forum?id=nTwb2vBLOV
[ "Zehao Dong", "Muhan Zhang", "Philip Payne", "Michael A Province", "Carlos Cruchaga", "Tianyu Zhao", "Fuhai Li", "Yixin Chen" ]
Poster
learning on graphs and other geometries & topologies
The expressivity of Graph Neural Networks (GNNs) has been studied broadly in recent years to reveal the design principles for more powerful GNNs. Graph canonization is known as a typical approach to distinguish non-isomorphic graphs, yet rarely adopted when developing expressive GNNs. This paper proposes to maximize th...
[ "Graph neural networks", "Graph canonization", "Stability" ]
null
9,229
2309.00738
title_snapshot
[ -0.01606675796210766, -0.02657218649983406, 0.004444750491529703, 0.046881429851055145, 0.03240280970931053, 0.02927512116730213, 0.03601134568452835, 0.006442246958613396, -0.010577226988971233, -0.04663076624274254, 0.0462639294564724, -0.023167768493294716, -0.0910857617855072, 0.005880...
Gradual Optimization Learning for Conformational Energy Minimization
https://openreview.net/forum?id=FMMF1a9ifL
[ "Artem Tsypin", "Leonid Anatolievich Ugadiarov", "Kuzma Khrabrov", "Alexander Telepov", "Egor Rumiantsev", "Alexey Skrynnik", "Aleksandr Panov", "Dmitry P. Vetrov", "Elena Tutubalina", "Artur Kadurin" ]
Poster
applications to physical sciences (physics, chemistry, biology, etc.)
Molecular conformation optimization is crucial to computer-aided drug discovery and materials design. Traditional energy minimization techniques rely on iterative optimization methods that use molecular forces calculated by a physical simulator (oracle) as anti-gradients. However, this is a computationally expensive ap...
[ "energy minimization", "conformational optimization", "geometry optimization" ]
We propose a data-efficient framework for conformational energy minimization with neural networks
9,224
2311.06295
title_snapshot
[ -0.015479356981813908, 0.003977349027991295, 0.012059963308274746, 0.03229477256536484, 0.05440262332558632, -0.00313603226095438, 0.018445691093802452, -0.002321593463420868, -0.001933555118739605, -0.03871912509202957, 0.02142464369535446, 0.021075304597616196, -0.05638939142227173, 0.00...
AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection
https://openreview.net/forum?id=buC4E91xZE
[ "Qihang Zhou", "Guansong Pang", "Yu Tian", "Shibo He", "Jiming Chen" ]
Poster
representation learning for computer vision, audio, language, and other modalities
Zero-shot anomaly detection (ZSAD) requires detection models trained using auxiliary data to detect anomalies without any training sample in a target dataset. It is a crucial task when training data is not accessible due to various concerns, e.g., data privacy, yet it is challenging since the models need to generalize ...
[ "Anomaly detection", "Zero-shot anomaly detection", "CLIP", "Industrial defect inspection" ]
We propose a novel approach, namely AnomalyCLIP, to adapt CLIP for accurate zero-shot anomaly detection.
9,222
2310.18961
title_snapshot
[ 0.007240860257297754, -0.016173506155610085, -0.02448737435042858, 0.04952886328101158, 0.04749135673046112, -0.010160885751247406, 0.031150704249739647, 0.0034091866109520197, -0.033539753407239914, -0.015604296699166298, -0.05279701575636864, 0.011703313328325748, -0.08205169439315796, 0...
Towards Faithful Explanations: Boosting Rationalization with Shortcuts Discovery
https://openreview.net/forum?id=uGtfk2OphU
[ "Linan Yue", "Qi Liu", "Yichao Du", "Li Wang", "Weibo Gao", "Yanqing An" ]
Poster
representation learning for computer vision, audio, language, and other modalities
The remarkable success in neural networks provokes the selective rationalization. It explains the prediction results by identifying a small subset of the inputs sufficient to support them. Since existing methods still suffer from adopting the shortcuts in data to compose rationales and limited large-scale annotated rat...
[ "Selective Rationalization", "Shortcut" ]
null
9,216
2403.07955
title_snapshot
[ -0.04196175932884216, -0.04155484214425087, -0.029909880831837654, 0.05383458361029625, 0.027852259576320648, 0.046177856624126434, 0.029774008318781853, -0.030364982783794403, -0.04535387083888054, -0.03539593145251274, -0.021743908524513245, 0.035830091685056686, -0.038910601288080215, -...
CORN: Contact-based Object Representation for Nonprehensile Manipulation of General Unseen Objects
https://openreview.net/forum?id=KTtEICH4TO
[ "Yoonyoung Cho", "Junhyek Han", "Yoontae Cho", "Beomjoon Kim" ]
Poster
applications to robotics, autonomy, planning
Nonprehensile manipulation is essential for manipulating objects that are too thin, large, or otherwise ungraspable in the wild. To sidestep the difficulty of contact modeling in conventional modeling-based approaches, reinforcement learning (RL) has recently emerged as a promising alternative. However, previous RL app...
[ "pretraining", "robotics", "manipulation", "object representation", "representation learning" ]
contact-based representation learning for nonprehensile robotic manipulation on objects with general geometry, with zero-shot real-world transfer.
9,206
2403.10760
title_snapshot
[ -0.021233784034848213, -0.0006242945091798902, -0.02503608725965023, 0.043408870697021484, 0.043300170451402664, 0.043824922293424606, -0.01094576995819807, 0.0053087687119841576, -0.027061792090535164, -0.05383054539561272, -0.034329600632190704, 0.006863035727292299, -0.08078441023826599, ...
An Intuitive Multi-Frequency Feature Representation for SO(3)-Equivariant Networks
https://openreview.net/forum?id=5JWAOLBxwp
[ "Dongwon Son", "Jaehyung Kim", "Sanghyeon Son", "Beomjoon Kim" ]
Poster
representation learning for computer vision, audio, language, and other modalities
The usage of 3D vision algorithms, such as shape reconstruction, remains limited because they require inputs to be at a fixed canonical rotation. Recently, a simple equivariant network, Vector Neuron (VN) has been proposed that can be easily used with the state-of-the-art 3D neural network (NN) architectures. However, ...
[ "Equivariant networks", "SO(3) Equivariance", "Fourier features" ]
We propose a frequency-based rotation equivariant feature representation for 3D data.
9,205
2405.04537
title_snapshot
[ 0.016153620555996895, -0.004348784685134888, 0.008651512674987316, 0.0016471713315695524, 0.010418830439448357, 0.04391166567802429, 0.009592808783054352, 0.01452427078038454, -0.05374327301979065, -0.06482110172510147, -0.005818979814648628, -0.024318881332874298, -0.06196344643831253, 0....
KoLA: Carefully Benchmarking World Knowledge of Large Language Models
https://openreview.net/forum?id=AqN23oqraW
[ "Jifan Yu", "Xiaozhi Wang", "Shangqing Tu", "Shulin Cao", "Daniel Zhang-Li", "Xin Lv", "Hao Peng", "Zijun Yao", "Xiaohan Zhang", "Hanming Li", "Chunyang Li", "Zheyuan Zhang", "Yushi Bai", "Yantao Liu", "Amy Xin", "Kaifeng Yun", "Linlu GONG", "Nianyi Lin", "Jianhui Chen", "Zhili...
Poster
datasets and benchmarks
The unprecedented performance of large language models (LLMs) necessitates improvements in evaluations. Rather than merely exploring the breadth of LLM abilities, we believe meticulous and thoughtful designs are essential to thorough, unbiased, and applicable evaluations. Given the importance of world knowledge to LLMs...
[ "Large Language Model", "World Knowledge", "Evolving Benchmark" ]
A carefully designed evolving benchmark for evaluating LLMs' world knowledge.
9,199
2306.09296
title_snapshot
[ -0.005490350537002087, -0.008638227358460426, -0.021806029602885246, 0.015115249902009964, 0.05535152181982994, 0.01104307733476162, 0.030813658609986305, 0.01802036724984646, -0.0243209395557642, 0.04231026768684387, -0.021940721198916435, 0.036305949091911316, -0.06189289689064026, -0.01...
Graph Parsing Networks
https://openreview.net/forum?id=hv3SklibkL
[ "Yunchong Song", "Siyuan Huang", "Xinbing Wang", "Chenghu Zhou", "Zhouhan Lin" ]
Poster
learning on graphs and other geometries & topologies
Graph pooling compresses graph information into a compact representation. State-of-the-art graph pooling methods follow a hierarchical approach, which reduces the graph size step-by-step. These methods must balance memory efficiency with preserving node information, depending on whether they use node dropping or node c...
[ "GNN", "graph pooling", "parsing" ]
In this paper, we propose an efficient parsing algorithm to infer the graph pooling structure and enable the model to learn personalized pooling structure for each individual graph.
9,196
2402.14393
title_snapshot
[ -0.0077302101999521255, -0.013876793906092644, 0.01468499843031168, 0.036350104957818985, 0.010527335107326508, 0.014487972483038902, 0.01972058415412903, 0.034563615918159485, -0.010012580081820488, -0.027285480871796608, 0.016318632289767265, -0.016860870644450188, -0.06835665553808212, ...
TESTAM: A Time-Enhanced Spatio-Temporal Attention Model with Mixture of Experts
https://openreview.net/forum?id=N0nTk5BSvO
[ "Hyunwook Lee", "Sungahn Ko" ]
Poster
unsupervised, self-supervised, semi-supervised, and supervised representation learning
Accurate traffic forecasting is challenging due to the complex dependency on road networks, various types of roads, and the abrupt speed change due to the events. Recent works mainly focus on dynamic spatial modeling with adaptive graph embedding or graph attention having less consideration for temporal characteristics...
[ "Traffic Prediction", "Deep Learning", "Spatio-Temporal data modeling" ]
We propose a novel mixture-of-experts model named TESTAM that enables in-situ modeling of the traffic data
9,189
2403.02600
title_snapshot
[ 0.01597568206489086, -0.034338634461164474, 0.017970934510231018, 0.009921878576278687, 0.019223932176828384, 0.03865485638380051, 0.030840985476970673, 0.040364403277635574, -0.01658823899924755, -0.03736557066440582, 0.015060276724398136, 0.016801465302705765, -0.0492691732943058, 0.0346...
Learning From Simplicial Data Based on Random Walks and 1D Convolutions
https://openreview.net/forum?id=OsGUnYOzii
[ "Florian Frantzen", "Michael T Schaub" ]
Poster
learning on graphs and other geometries & topologies
Triggered by limitations of graph-based deep learning methods in terms of computational expressivity and model flexibility, recent years have seen a surge of interest in computational models that operate on higher-order topological domains such as hypergraphs and simplicial complexes. While the increased expressivity o...
[ "simplicial complex", "simplicial neural network", "random walks" ]
null
9,185
2404.03434
title_snapshot
[ -0.004623337648808956, -0.037601593881845474, 0.010211730375885963, 0.058403052389621735, 0.007949823513627052, 0.023430993780493736, 0.009594488888978958, 0.025343045592308044, 0.018567826598882675, -0.05670231953263283, 0.011705897748470306, -0.03297681361436844, -0.08074546605348587, 0....
LUM-ViT: Learnable Under-sampling Mask Vision Transformer for Bandwidth Limited Optical Signal Acquisition
https://openreview.net/forum?id=wkbeqr5XhC
[ "Lingfeng Liu", "Dong Ni", "Hangjie Yuan" ]
Poster
general machine learning (i.e., none of the above)
Bandwidth constraints during signal acquisition frequently impede real-time detection applications. Hyperspectral data is a notable example, whose vast volume compromises real-time hyperspectral detection. To tackle this hurdle, we introduce a novel approach leveraging pre-acquisition modulation to reduce the acquisiti...
[ "hyperspectral imaging", "optical modulation", "real-time detection", "vision transformer", "pre-acquisition modulation", "learnable mask", "weight binarization" ]
null
9,181
2403.01412
title_snapshot
[ 0.017940182238817215, -0.029117709025740623, 0.014595971442759037, 0.016759345307946205, 0.038791507482528687, 0.01592128537595272, 0.005847859662026167, -0.0012538720620796084, -0.00969577208161354, -0.06602685153484344, -0.0009684857213869691, 0.05052831768989563, -0.04274393990635872, 0...
Social-Transmotion: Promptable Human Trajectory Prediction
https://openreview.net/forum?id=SQpnEfv9WH
[ "Saeed Saadatnejad", "Yang Gao", "Kaouther Messaoud", "Alexandre Alahi" ]
Poster
applications to robotics, autonomy, planning
Accurate human trajectory prediction is crucial for applications such as autonomous vehicles, robotics, and surveillance systems. Yet, existing models often fail to fully leverage the non-verbal social cues human subconsciously communicate when navigating the space. To address this, we introduce *Social-Transmotion*, a...
[ "human trajectory prediction", "robot navigation", "autonomous driving", "attention mechanism" ]
We propose a generic Transformer-based model that integrates diverse visual cues as prompts, powered by masking technique to enhance human trajectory prediction.
9,161
2312.16168
title_snapshot
[ 0.018512999638915062, -0.03970286250114441, -0.00012769195018336177, 0.02089161053299904, 0.0005527804605662823, -0.012500859797000885, 0.05488057807087898, 0.0340292826294899, -0.020628076046705246, -0.044241759926080704, -0.05694842338562012, -0.02115008421242237, -0.06498681753873825, -...
Robust Classification via Regression for Learning with Noisy Labels
https://openreview.net/forum?id=wfgZc3IMqo
[ "Erik Englesson", "Hossein Azizpour" ]
Poster
unsupervised, self-supervised, semi-supervised, and supervised representation learning
Deep neural networks and large-scale datasets have revolutionized the field of machine learning. However, these large networks are susceptible to overfitting to label noise, resulting in reduced generalization. To address this challenge, two promising approaches have emerged: i) loss reweighting, which reduces the infl...
[ "label noise", "noisy labels", "robustness", "Gaussian noise", "classification", "log-ratio transform", "compositional data analysis" ]
null
9,158
null
null
[ 0.007654992397874594, -0.012546625919640064, -0.02607038989663124, 0.037611931562423706, 0.02523856610059738, 0.049583956599235535, 0.010495726019144058, -0.024648746475577354, -0.027882399037480354, -0.06036657467484474, -0.024021822959184647, 0.011940705589950085, -0.0753832757472992, 0....
Generalization error of spectral algorithms
https://openreview.net/forum?id=3SJE1WLB4M
[ "Maksim Velikanov", "Maxim Panov", "Dmitry Yarotsky" ]
Spotlight
learning theory
The asymptotically precise estimation of the generalization of kernel methods has recently received attention due to the parallels between neural networks and their associated kernels. However, prior works derive such estimates for training by kernel ridge regression (KRR), whereas neural networks are typically trained...
[ "gradient descent", "kernel ridge regression", "optimal algorithm", "generalization", "asymptotic error rates", "power-laws" ]
null
9,152
2403.11696
title_snapshot
[ -0.03708674758672714, -0.015557538717985153, 0.02358335070312023, 0.024285510182380676, 0.029260801151394844, 0.043063919991254807, 0.05715901777148247, -0.009645843878388405, -0.024643635377287865, -0.03209058567881584, -0.018496034666895866, 0.028681425377726555, -0.06809186190366745, 0....
Learning to Reject with a Fixed Predictor: Application to Decontextualization
https://openreview.net/forum?id=dCHbFDsCZz
[ "Christopher Mohri", "Daniel Andor", "Eunsol Choi", "Michael Collins", "Anqi Mao", "Yutao Zhong" ]
Poster
general machine learning (i.e., none of the above)
We study the problem of classification with a reject option for a fixed predictor, crucial to natural language processing. We introduce a new problem formulation for this scenario, and an algorithm minimizing a new surrogate loss function. We provide a complete theoretical analysis of the surrogate loss function with a...
[ "Rejection", "abstention", "loss function", "consistency", "learning theory", "decontextualization", "natural language processing" ]
We introduce a new loss function for learning a rejector with a fixed predictor, and focus on the decontextualization task.
9,141
2301.09044
title_snapshot
[ -0.0023585702292621136, -0.014705046080052853, -0.025719713419675827, 0.04118062183260918, 0.03664138913154602, 0.05527365580201149, 0.022560138255357742, 0.014385717920958996, 0.005115728825330734, -0.02304106391966343, -0.02516329661011696, 0.028997089713811874, -0.07568001747131348, 0.0...
Dynamics-Informed Protein Design with Structure Conditioning
https://openreview.net/forum?id=jZPqf2G9Sw
[ "Urszula Julia Komorowska", "Simon V Mathis", "Kieran Didi", "Francisco Vargas", "Pietro Lio", "Mateja Jamnik" ]
Poster
applications to physical sciences (physics, chemistry, biology, etc.)
Current protein generative models are able to design novel backbones with desired shapes or functional motifs. However, despite the importance of a protein’s dynamical properties for its function, conditioning on dynamical properties remains elusive. We present a new approach to protein generative modeling by leveragin...
[ "Diffusion Models", "Generative Modeling", "Protein Design", "Normal Mode Analysis" ]
null
9,128
null
null
[ -0.02976337820291519, -0.010502273216843605, -0.007492063567042351, 0.02982894703745842, 0.03016866371035576, -0.0036579701118171215, 0.02563508413732052, -0.027722373604774475, -0.001792095834389329, -0.0543234683573246, 0.03345227986574173, -0.012677071616053581, -0.06713400036096573, 0....
Partitioning Message Passing for Graph Fraud Detection
https://openreview.net/forum?id=tEgrUrUuwA
[ "Wei Zhuo", "Zemin Liu", "Bryan Hooi", "Bingsheng He", "Guang Tan", "Rizal Fathony", "Jia Chen" ]
Poster
learning on graphs and other geometries & topologies
Label imbalance and homophily-heterophily mixture are the fundamental problems encountered when applying Graph Neural Networks (GNNs) to Graph Fraud Detection (GFD) tasks. Existing GNN-based GFD models are designed to augment graph structure to accommodate the inductive bias of GNNs towards homophily, by excluding hete...
[ "Graph Neural Networks", "Fraud Detection" ]
Applyiing GNNs on fraud detection by distinguishing neighbors during message passing.
9,114
2412.00020
title_snapshot
[ -0.015180970542132854, -0.03004416823387146, -0.00014610648213420063, 0.0534706674516201, 0.02625679038465023, 0.030423501506447792, 0.029557548463344574, 0.00843799114227295, -0.03424273431301117, -0.029521716758608818, 0.034313127398490906, -0.0070055145770311356, -0.09498010575771332, 0...
Self-contradictory Hallucinations of Large Language Models: Evaluation, Detection and Mitigation
https://openreview.net/forum?id=EmQSOi1X2f
[ "Niels Mündler", "Jingxuan He", "Slobodan Jenko", "Martin Vechev" ]
Poster
societal considerations including fairness, safety, privacy
Large language models (large LMs) are susceptible to producing text that contains hallucinated content. An important instance of this problem is self-contradiction, where the LM generates two contradictory sentences within the same context. In this work, we present a comprehensive investigation into self-contradiction ...
[ "language model", "hallucination", "trustworthy artificial intelligence", "reasoning" ]
We present a comprehensive analysis showing that state-of-the-art LLMs frequently produce self-contradictory hallucinations. We then design prompting methods that effectively detect and mitigate self-contradictions.
9,113
2305.15852
title_snapshot
[ -0.011516749858856201, -0.025782722979784012, -0.013178627006709576, 0.05657072365283966, 0.023291703313589096, -0.006964897271245718, 0.0577579103410244, 0.04383806139230728, -0.02117857150733471, -0.01919715665280819, -0.04526984319090843, 0.05777693912386894, -0.07427429407835007, -0.00...
Debiased Collaborative Filtering with Kernel-Based Causal Balancing
https://openreview.net/forum?id=Ffjc8ApSbt
[ "Haoxuan Li", "Chunyuan Zheng", "Yanghao Xiao", "Peng Wu", "Zhi Geng", "Xu Chen", "Peng Cui" ]
Spotlight
general machine learning (i.e., none of the above)
Collaborative filtering builds personalized models from the collected user feedback. However, the collected data is observational rather than experimental, leading to various biases in the data, which can significantly affect the learned model. To address this issue, many studies have focused on propensity-based method...
[ "Recommender System", "Causal Inference", "Bias", "Debias", "Balancing" ]
null
9,073
2404.19596
title_snapshot
[ 0.025419780984520912, -0.03640938177704811, 0.028926391154527664, 0.018276844173669815, 0.03656871244311333, 0.025370893999934196, 0.019817065447568893, -0.01657688245177269, -0.012147109024226665, -0.0559467151761055, -0.0021596152801066637, 0.032072242349386215, -0.08304335176944733, 0.0...
Symmetric Neural-Collapse Representations with Supervised Contrastive Loss: The Impact of ReLU and Batching
https://openreview.net/forum?id=AyXIDfvYg8
[ "Ganesh Ramachandra Kini", "Vala Vakilian", "Tina Behnia", "Jaidev Gill", "Christos Thrampoulidis" ]
Poster
unsupervised, self-supervised, semi-supervised, and supervised representation learning
Supervised contrastive loss (SCL) is a competitive and often superior alternative to the cross-entropy loss for classification. While prior studies have demonstrated that both losses yield symmetric training representations under balanced data, this symmetry breaks under class imbalances. This paper presents an intrigu...
[ "Supervised contrastive learning", "neural collapse", "implicit bias", "class imbalance" ]
null
9,066
2306.07960
title_snapshot
[ -0.015492595732212067, -0.015959473326802254, 0.0020398942288011312, 0.025583665817975998, 0.0034498898312449455, 0.03848980367183685, 0.02236349694430828, -0.0024522251915186644, -0.05987273156642914, -0.036690160632133484, -0.002380494261160493, -0.03404854238033295, -0.0697200819849968, ...
Manipulating dropout reveals an optimal balance of efficiency and robustness in biological and machine visual systems
https://openreview.net/forum?id=ADDCErFzev
[ "Jacob S. Prince", "Gabriel Fajardo", "George A. Alvarez", "Talia Konkle" ]
Poster
applications to neuroscience & cognitive science
According to the efficient coding hypothesis, neural populations encode information optimally when representations are high-dimensional and uncorrelated. However, such codes may carry a cost in terms of generalization and robustness. Past empirical studies of early visual cortex (V1) in rodents have suggested that this...
[ "Efficient coding", "object representation", "dropout", "robustness", "human fMRI", "occipitotemporal cortex", "cognitive neuroscience", "distributed coding" ]
null
9,065
null
null
[ -0.01782986707985401, 0.006622664630413055, 0.007570574525743723, 0.03355454280972481, 0.018320530652999878, 0.008736545220017433, 0.049170464277267456, 0.03632017597556114, -0.04536814242601395, -0.06825223565101624, -0.03937917202711105, -0.01697658747434616, -0.07647861540317535, -0.011...
DDMI: Domain-agnostic Latent Diffusion Models for Synthesizing High-Quality Implicit Neural Representations
https://openreview.net/forum?id=327tbF3S65
[ "Dogyun Park", "Sihyeon Kim", "Sojin Lee", "Hyunwoo J. Kim" ]
Poster
generative models
Recent studies have introduced a new class of generative models for synthesizing implicit neural representations (INRs) that capture arbitrary continuous signals in various domains. These models opened the door for domain-agnostic generative models, but they often fail to achieve high-quality generation. We observed th...
[ "Implicit neural representation", "generative model", "domain agnostic", "diffusion model" ]
We propose a latent diffusion model that generates hierarchically decomposed positional embeddings of Implicit neural representations, enabling high-quality generation on various data domains.
9,058
2401.12517
title_snapshot
[ 0.0015107319923117757, -0.020144982263445854, -0.009731330908834934, 0.055767662823200226, 0.05077431723475456, 0.03490469977259636, 0.015939394012093544, -0.005261501297354698, -0.005571797490119934, -0.042657334357500076, 0.005277602467685938, -0.02570209465920925, -0.038372088223695755, ...
Bayesian Coreset Optimization for Personalized Federated Learning
https://openreview.net/forum?id=uz7d2N2zul
[ "Prateek Chanda", "Shrey Modi", "Ganesh Ramakrishnan" ]
Poster
general machine learning (i.e., none of the above)
In a distributed machine learning setting like Federated Learning where there are multiple clients involved which update their individual weights to a single central server, often training on the entire individual client's dataset for each client becomes cumbersome. To address this issue we propose CORESET-PFEDBAYES : ...
[ "federated learning", "personalized federated learning", "bayesian coreset", "submodularity", "variational inference", "coresets", "optimization" ]
The paper deals with utilizing a bayesian coreset on individual client's data in a federated learning setting that takes into account personalization at each client's side.
9,034
2511.01800
title_snapshot
[ -0.017540715634822845, -0.037643007934093475, 0.006482667755335569, 0.043998703360557556, 0.05224296450614929, 0.043822623789310455, 0.012661303393542767, -0.021877363324165344, 0.0032320937607437372, -0.058917198330163956, -0.004543339367955923, -0.00460267486050725, -0.028018053621053696, ...
In-context Autoencoder for Context Compression in a Large Language Model
https://openreview.net/forum?id=uREj4ZuGJE
[ "Tao Ge", "Hu Jing", "Lei Wang", "Xun Wang", "Si-Qing Chen", "Furu Wei" ]
Poster
representation learning for computer vision, audio, language, and other modalities
We propose the In-context Autoencoder (ICAE), leveraging the power of a large language model (LLM) to compress a long context into short compact memory slots that can be directly conditioned on by the LLM for various purposes. ICAE is first pretrained using both autoencoding and language modeling objectives on massive ...
[ "large language model", "context compression", "in-context autoencoder", "pretraining", "fine-tuning", "Llama", "GPT", "memorization" ]
A paper proposing a novel approach called In-context Autoencoder (ICAE) for LLM context compression
9,031
2307.06945
title_snapshot
[ 0.001845542574301362, 0.006097600795328617, -0.040349286049604416, 0.03521633520722389, 0.03832735866308212, 0.0520038902759552, 0.032748959958553314, 0.046779461205005646, -0.01397815253585577, -0.018891045823693275, -0.03845401853322983, 0.014703861437737942, -0.028778821229934692, -0.00...
Mitigating Hallucination in Large Multi-Modal Models via Robust Instruction Tuning
https://openreview.net/forum?id=J44HfH4JCg
[ "Fuxiao Liu", "Kevin Lin", "Linjie Li", "Jianfeng Wang", "Yaser Yacoob", "Lijuan Wang" ]
Poster
datasets and benchmarks
Despite the promising progress in multi-modal tasks, current large multi-modal models (LMMs) are prone to hallucinating inconsistent descriptions with respect to the associated image and human instructions. This paper addresses this issue by introducing the first large and diverse visual instruction tuning dataset, nam...
[ "instruction tuning", "multimodal large language model", "hallucination", "datasets" ]
null
9,027
2306.14565
title_snapshot
[ 0.011697652749717236, 0.007265315856784582, -0.005445276852697134, 0.02818222902715206, 0.018674831837415695, -0.004033592063933611, 0.048587918281555176, 0.030370205640792847, -0.007768841460347176, -0.03519640490412712, -0.0270729660987854, 0.027050962671637535, -0.08734658360481262, 0.0...
Multimarginal Generative Modeling with Stochastic Interpolants
https://openreview.net/forum?id=FHqAzWl2wE
[ "Michael Samuel Albergo", "Nicholas Matthew Boffi", "Michael Lindsey", "Eric Vanden-Eijnden" ]
Poster
generative models
Given a set of $K$ probability densities, we consider the multimarginal generative modeling problem of learning a joint distribution that recovers these densities as marginals. The structure of this joint distribution should identify multi-way correspondences among the prescribed marginals. We formalize an approach to ...
[ "multi-marginal", "unsupervised learning", "generative modeling", "measure transport", "optimal transport" ]
We introduce a method to generalize flow-based and diffusion based generative models to map between K distributions instead of two, revealing multiway-correspondences between densities.
9,021
2310.03695
title_snapshot
[ -0.019607307389378548, -0.028552260249853134, -0.003435323480516672, 0.06121153384447098, 0.018672902137041092, 0.03590376675128937, -0.0016992584569379687, 0.0020226328633725643, -0.020892340689897537, -0.06401150673627853, -0.005131365265697241, -0.011214197613298893, -0.06407607346773148,...
Proving Test Set Contamination in Black-Box Language Models
https://openreview.net/forum?id=KS8mIvetg2
[ "Yonatan Oren", "Nicole Meister", "Niladri S. Chatterji", "Faisal Ladhak", "Tatsunori Hashimoto" ]
Oral
societal considerations including fairness, safety, privacy
Large language models are trained on vast amounts of internet data, prompting concerns that they have memorized public benchmarks. Detecting this type of contamination is challenging because the pretraining data used by proprietary models are often not publicly accessible. We propose a procedure for detecting test set...
[ "language modeling", "memorization", "dataset contamination" ]
null
9,019
2310.17623
title_snapshot
[ -0.02294420823454857, -0.0004916504840366542, -0.05435055494308472, 0.0842687264084816, 0.03656534478068352, 0.024771148338913918, 0.03006507270038128, 0.014668131247162819, -0.016570713371038437, -0.0011257784208282828, -0.015860166400671005, 0.05113253369927406, -0.06834419816732407, -0....
Rethinking Channel Dependence for Multivariate Time Series Forecasting: Learning from Leading Indicators
https://openreview.net/forum?id=JiTVtCUOpS
[ "Lifan Zhao", "Yanyan Shen" ]
Poster
general machine learning (i.e., none of the above)
Recently, channel-independent methods have achieved state-of-the-art performance in multivariate time series (MTS) forecasting. Despite reducing overfitting risks, these methods miss potential opportunities in utilizing channel dependence for accurate predictions. We argue that there exist locally stationary lead-lag r...
[ "Multivariate time series forecasting", "channel dependence", "lead-lag relationships", "distribution shift" ]
We rethink the distribution shift in multivariate time series, where channel dependence varies over time. We propose LIFT to dynamically select and leverage leading indicators, which emprically improves SOTA forecasting methods by 5.5% in average.
9,003
2401.17548
title_snapshot
[ -0.02887018956243992, -0.048905909061431885, -0.008601301349699497, -0.011274679563939571, 0.04410098120570183, 0.05458495393395424, 0.04934972524642944, -0.013428439386188984, -0.03496110811829567, -0.041954927146434784, -0.012202408164739609, 0.02338135614991188, -0.06755620241165161, 0....
RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval
https://openreview.net/forum?id=GN921JHCRw
[ "Parth Sarthi", "Salman Abdullah", "Aditi Tuli", "Shubh Khanna", "Anna Goldie", "Christopher D Manning" ]
Poster
generative models
Retrieval-augmented language models can better adapt to changes in world state and incorporate long-tail knowledge. However, most existing methods retrieve only short contiguous chunks from a retrieval corpus, limiting holistic understanding of the overall document context. We introduce the novel approach of recursive...
[ "Retrieval Augmented Language Models", "Information Retrieval", "summarization", "QA" ]
RAPTOR improves LLM QA performance by constructing a hierarchical summarization tree for information retrieval, outperforming existing retrieval methods across various metrics and datasets.
8,997
2401.18059
title_snapshot
[ -0.028084367513656616, -0.03808955103158951, -0.0069338432513177395, 0.058946944773197174, 0.03572374954819679, -0.010479122400283813, 0.02182866632938385, 0.04068976640701294, -0.014082591980695724, 0.013376738876104355, -0.04412154108285904, 0.026203574612736702, -0.0667194277048111, -0....
Fair and Efficient Contribution Valuation for Vertical Federated Learning
https://openreview.net/forum?id=sLQb8q0sUi
[ "Zhenan Fan", "Huang Fang", "Xinglu Wang", "Zirui Zhou", "Jian Pei", "Michael Friedlander", "Yong Zhang" ]
Poster
societal considerations including fairness, safety, privacy
Federated learning is an emerging technology for training machine learning models across decentralized data sources without sharing data. Vertical federated learning, also known as feature-based federated learning, applies to scenarios where data sources have the same sample IDs but different feature sets. To ensure fa...
[ "Vertical federated learning", "Contribution valuation", "Fairness" ]
null
8,996
2201.02658
title_snapshot
[ -0.007843809202313423, -0.04759475216269493, 0.024263475090265274, 0.03280773386359215, 0.02604135125875473, 0.028319241479039192, 0.018956266343593597, -0.003370399819687009, -0.034652821719646454, -0.0583292655646801, 0.01480644941329956, -0.022529026493430138, -0.06296131014823914, 0.01...
In-Context Learning through the Bayesian Prism
https://openreview.net/forum?id=HX5ujdsSon
[ "Madhur Panwar", "Kabir Ahuja", "Navin Goyal" ]
Poster
transfer learning, meta learning, and lifelong learning
In-context learning (ICL) is one of the surprising and useful features of large language models and subject of intense research. Recently, stylized meta-learning-like ICL setups have been devised that train transformers on sequences of input-output pairs $(x, f(x))$. The function $f$ comes from a function class and gen...
[ "In-context Learning", "Transformers", "Inductive Biases", "Meta Learning", "Language Modelling", "Bayesian Inference" ]
null
8,985
2306.04891
title_snapshot
[ -0.0128860492259264, 0.020753707736730576, -0.016711251810193062, 0.03542608395218849, 0.01987953670322895, 0.012075168080627918, 0.03337359055876732, 0.026288079097867012, -0.032102085649967194, 0.013129864819347858, -0.03737231716513634, 0.06185484677553177, -0.05695034936070442, -0.0228...
RingAttention with Blockwise Transformers for Near-Infinite Context
https://openreview.net/forum?id=WsRHpHH4s0
[ "Hao Liu", "Matei Zaharia", "Pieter Abbeel" ]
Poster
unsupervised, self-supervised, semi-supervised, and supervised representation learning
Transformers have emerged as the architecture of choice for many state-of-the-art AI models, showcasing exceptional performance across a wide range of AI applications. However, the memory demands imposed by Transformers limit their ability to handle long sequences, thereby posing challenges in utilizing videos, actions...
[ "Language Model", "Large Context", "Transformers", "Long Context Model", "Memory Efficiency" ]
We present an efficient method of computing the standard transformer architecture, enabling effective processing of long contextual information.
8,983
2310.01889
title_judge
[ -0.0008233634871430695, -0.02736123837530613, -0.0026071430183947086, 0.027651285752654076, 0.027741095051169395, 0.012516405433416367, 0.02917931042611599, 0.004430645611137152, -0.03680356964468956, -0.011788606643676758, -0.04366142302751541, -0.022835751995444298, -0.06083191558718681, ...
Chain of Hindsight aligns Language Models with Feedback
https://openreview.net/forum?id=6xfe4IVcOu
[ "Hao Liu", "Carmelo Sferrazza", "Pieter Abbeel" ]
Poster
reinforcement learning
Learning from human preferences is important for language models to match human needs and to align with human and social values. Prior works have achieved remarkable successes by learning from human feedback to understand and follow instructions. Nonetheless, these methods are either founded on hand-picked model gener...
[ "Reinforcement Learning", "Reinforcement Learning from Human Feedback", "RLHF" ]
We present a method for learning from human feedback that's simpler than RLHF
8,976
2302.02676
title_snapshot
[ -0.016200540587306023, -0.0133229810744524, -0.0013959624338895082, 0.021960871294140816, 0.04265383630990982, 0.029652118682861328, 0.036909691989421844, 0.04218015819787979, -0.01908288151025772, -0.010578646324574947, -0.026119671761989594, 0.07389149814844131, -0.05307700112462044, -0....
GraphChef: Decision-Tree Recipes to Explain Graph Neural Networks
https://openreview.net/forum?id=IjMUGuUmBI
[ "Peter Müller", "Lukas Faber", "Karolis Martinkus", "Roger Wattenhofer" ]
Poster
visualization or interpretation of learned representations
We propose a new self-explainable Graph Neural Network (GNN) model: GraphChef. GraphChef integrates decision trees into the GNN message passing framework. Given a dataset, GraphChef returns a set of rules (a recipe) that explains each class in the dataset unlike existing GNNs and explanation methods that reason on indi...
[ "Graph Neural Networks", "GNN", "Explainability", "Decision Trees" ]
GraphChef integrate Decision Trees into Graph Neural Networks to allow explaining the full decision process.
8,971
null
null
[ -0.03824394568800926, -0.03040553629398346, -0.02923457883298397, 0.04446413367986679, 0.030412863940000534, 0.032183799892663956, -0.009437767788767815, 0.010258176364004612, -0.018317122012376785, -0.04904341697692871, 0.000421699631260708, 0.013033793307840824, -0.046178556978702545, 0....
Safe Collaborative Filtering
https://openreview.net/forum?id=yarUvgEXq3
[ "Riku Togashi", "Tatsushi Oka", "Naoto Ohsaka", "Tetsuro Morimura" ]
Poster
general machine learning (i.e., none of the above)
Excellent tail performance is crucial for modern machine learning tasks, such as algorithmic fairness, class imbalance, and risk-sensitive decision making, as it ensures the effective handling of challenging samples within a dataset. Tail performance is also a vital determinant of success for personalized recommender s...
[ "recommender systems", "collaborative filtering", "scalability" ]
null
8,970
2306.05292
title_snapshot
[ 0.026897773146629333, -0.02315434068441391, 0.02918044663965702, 0.019358398392796516, 0.07579630613327026, -0.018848463892936707, 0.04576868191361427, -0.00825085025280714, -0.010849758051335812, -0.0452670156955719, -0.02962655946612358, 0.017351757735013962, -0.05180773511528969, -0.004...
On Representation Complexity of Model-based and Model-free Reinforcement Learning
https://openreview.net/forum?id=3K3s9qxSn7
[ "Hanlin Zhu", "Baihe Huang", "Stuart Russell" ]
Poster
learning theory
We study the representation complexity of model-based and model-free reinforcement learning (RL) in the context of circuit complexity. We prove theoretically that there exists a broad class of MDPs such that their underlying transition and reward functions can be represented by constant depth circuits with polynomial s...
[ "model-based and model-free RL", "representation complexity", "circuit complexity", "approximation error" ]
We study representation complexity of model-based and model-free RL through circuit complexity to provide unique insights into sample efficiency of model-based RL.
8,968
2310.01706
title_snapshot
[ -0.04927462339401245, -0.0062835062853991985, -0.0047943079844117165, 0.050318337976932526, 0.051472023129463196, 0.01690002717077732, 0.00744330370798707, -0.0033518101554363966, -0.03218875825405121, -0.009630223736166954, 0.002437263960018754, 0.02867736853659153, -0.06386642903089523, ...
Addressing Loss of Plasticity and Catastrophic Forgetting in Continual Learning
https://openreview.net/forum?id=sKPzAXoylB
[ "Mohamed Elsayed", "A. Rupam Mahmood" ]
Poster
transfer learning, meta learning, and lifelong learning
Deep representation learning methods struggle with continual learning, suffering from both catastrophic forgetting of useful units and loss of plasticity, often due to rigid and unuseful units. While many methods address these two issues separately, only a few currently deal with both simultaneously. In this paper, we ...
[ "catastrophic forgetting", "loss of plasticity", "plasticity", "stability", "continual learning", "streaming learning", "online learning", "incremental learning" ]
null
8,965
2404.00781
title_snapshot
[ -0.038363005965948105, -0.034640535712242126, -0.003993376158177853, 0.036814406514167786, 0.02639305219054222, 0.04206480085849762, 0.01803930103778839, 0.02440030500292778, -0.054143521934747696, -0.02579873614013195, -0.002535837469622493, 0.007489477749913931, -0.06740778684616089, -0....
A Good Learner can Teach Better: Teacher-Student Collaborative Knowledge Distillation
https://openreview.net/forum?id=Ixi4j6LtdX
[ "Ayan Sengupta", "Shantanu Dixit", "Md Shad Akhtar", "Tanmoy Chakraborty" ]
Poster
transfer learning, meta learning, and lifelong learning
Knowledge distillation (KD) is a technique used to transfer knowledge from a larger ''teacher'' model into a smaller ''student'' model. Recent advancements in meta-learning-based knowledge distillation (MetaKD) emphasize that the fine-tuning of teacher models should be aware of the student's need to achieve better know...
[ "Knowledge Distillation", "Meta-Knowledge Distillation", "Policy-driven Knowledge Distillation", "Large Language Models" ]
The paper introduces collaborative joint loss and curriculum learning for meta-teacher knowledge distillation
8,958
null
null
[ -0.018022816628217697, -0.024362346157431602, 0.0017430695006623864, 0.07432534545660019, 0.06194327771663666, -0.02464231289923191, 0.0462917685508728, -0.004169303923845291, -0.03083052858710289, 0.010140984319150448, -0.042350880801677704, 0.012524588964879513, -0.032330308109521866, 0....
Diagnosing Transformers: Illuminating Feature Spaces for Clinical Decision-Making
https://openreview.net/forum?id=k581sTMyPt
[ "Aliyah R. Hsu", "Yeshwanth Cherapanamjeri", "Briton Park", "Tristan Naumann", "Anobel Odisho", "Bin Yu" ]
Poster
visualization or interpretation of learned representations
Pre-trained transformers are often fine-tuned to aid clinical decision-making using limited clinical notes. Model interpretability is crucial, especially in high-stakes domains like medicine, to establish trust and ensure safety, which requires human engagement. We introduce SUFO, a systematic framework that enhances i...
[ "fine-tuning", "transformer-based language models", "feature analysis", "interpretation", "clinical classification" ]
null
8,956
2305.17588
title_snapshot
[ -0.02523307502269745, -0.042041558772325516, -0.012514008209109306, 0.00452608335763216, 0.06782407313585281, -0.0028522212523967028, 0.02807779237627983, -0.03189451992511749, 0.023583458736538887, -0.04348624497652054, -0.014005664736032486, 0.07299472391605377, -0.060723260045051575, 0....
Mechanistically analyzing the effects of fine-tuning on procedurally defined tasks
https://openreview.net/forum?id=A0HKeKl4Nl
[ "Samyak Jain", "Robert Kirk", "Ekdeep Singh Lubana", "Robert P. Dick", "Hidenori Tanaka", "Tim Rocktäschel", "Edward Grefenstette", "David Krueger" ]
Poster
visualization or interpretation of learned representations
Fine-tuning large pre-trained models has become the de facto strategy for developing both task-specific and general-purpose machine learning systems, including developing models that are safe to deploy. Despite its clear importance, there has been minimal work that explains how fine-tuning alters the underlying capabil...
[ "Fine-Tuning", "Interpretability", "Mechanisms" ]
We demonstrate that fine-tuning models rarely alters their underlying capabilities.
8,951
2311.12786
title_snapshot
[ -0.00892076175659895, -0.03230709955096245, -0.023379942402243614, 0.014001891948282719, 0.07587410509586334, 0.01001657359302044, 0.04164785146713257, -0.010055135935544968, -0.03506295010447502, -0.0431143157184124, -0.002066337503492832, 0.029455015435814857, -0.030989952385425568, -0.0...
RepoBench: Benchmarking Repository-Level Code Auto-Completion Systems
https://openreview.net/forum?id=pPjZIOuQuF
[ "Tianyang Liu", "Canwen Xu", "Julian McAuley" ]
Poster
datasets and benchmarks
Large Language Models (LLMs) have greatly advanced code auto-completion systems, with a potential for substantial productivity enhancements for developers. However, current benchmarks mainly focus on single-file tasks, leaving an assessment gap for more complex, real-world, multi-file programming scenarios. To fill thi...
[ "large language model", "code completion", "benchmark" ]
We introduce RepoBench, a comprehensive benchmark for evaluating repository-level code auto-completion systems
8,936
2306.03091
title_snapshot
[ -0.02862677350640297, -0.06694172322750092, -0.0005661206669174135, 0.024119915440678596, 0.029531128704547882, 0.01935969479382038, 0.03747379779815674, 0.03156756982207298, -0.05049164593219757, -0.05522480234503746, -0.027455320581793785, -0.007860885001718998, -0.03932227939367294, -0....
Seeking Neural Nuggets: Knowledge Transfer in Large Language Models from a Parametric Perspective
https://openreview.net/forum?id=mIEHIcHGOo
[ "Ming Zhong", "Chenxin An", "Weizhu Chen", "Jiawei Han", "Pengcheng He" ]
Poster
representation learning for computer vision, audio, language, and other modalities
Large Language Models (LLMs) inherently encode a wealth of knowledge within their parameters through pre-training on extensive corpora. While prior research has delved into operations on these parameters to manipulate the underlying implicit knowledge — encompassing detection, editing, and merging — there remains an am...
[ "Parametric Knowledge Transfer", "Large Language Model" ]
In this paper, we provide empirical evidence that parametric knowledge are transferable between large language models accross varying scales.
8,932
2310.11451
title_snapshot
[ -0.027943428605794907, -0.025494633242487907, -0.016240498051047325, 0.022230330854654312, 0.05116257071495056, 0.03690645843744278, 0.041253406554460526, 0.023194124922156334, -0.01525137946009636, 0.006928251590579748, -0.022543855011463165, 0.030451903119683266, -0.044833406805992126, 0...
Harnessing Explanations: LLM-to-LM Interpreter for Enhanced Text-Attributed Graph Representation Learning
https://openreview.net/forum?id=RXFVcynVe1
[ "Xiaoxin He", "Xavier Bresson", "Thomas Laurent", "Adam Perold", "Yann LeCun", "Bryan Hooi" ]
Poster
learning on graphs and other geometries & topologies
Representation learning on text-attributed graphs (TAGs) has become a critical research problem in recent years. A typical example of a TAG is a paper citation graph, where the text of each paper serves as node attributes. Initial graph neural network (GNN) pipelines handled these text attributes by transforming them i...
[ "large language models (LLM)", "feature learning", "text attributed graphs (TAG)", "graph neural networks (GNN)" ]
We propose the first framework that leverages LLMs to enhance representation learning on text-attributed graphs, achieving SOTA results on four benchmark datasets.
8,930
2305.19523
title_snapshot
[ -0.0036593638360500336, -0.022927701473236084, -0.016103394329547882, 0.048348262906074524, 0.03673917055130005, -0.004856843966990709, -0.002440623240545392, 0.021300513297319412, -0.0007352242828346789, -0.007171661127358675, -0.020632246509194374, 0.026297302916646004, -0.0655625835061073...
The Effective Horizon Explains Deep RL Performance in Stochastic Environments
https://openreview.net/forum?id=5ES5Hdlbxw
[ "Cassidy Laidlaw", "Banghua Zhu", "Stuart Russell", "Anca Dragan" ]
Spotlight
reinforcement learning
Reinforcement learning (RL) theory has largely focused on proving minimax sample complexity bounds. These require strategic exploration algorithms that use relatively limited function classes for representing the policy or value function. Our goal is to explain why deep RL algorithms often perform well in practice, des...
[ "reinforcement learning", "effective horizon", "RL theory", "theory of reinforcement learning", "instance-dependent bounds", "empirical validation of theory" ]
We extend the recently proposed "effective horizon" property to stochastic MDPs and show theoretically and empirically that it can explain the performance of deep RL algorithms.
8,925
2312.08369
title_snapshot
[ -0.0320478230714798, -0.029261328279972076, 0.004281132481992245, 0.05770476907491684, 0.058060165494680405, 0.031277649104595184, 0.021932845935225487, 0.008290723897516727, -0.03545078635215759, -0.022893091663718224, -0.021227650344371796, 0.011919436976313591, -0.07415973395109177, -0....
SuRe: Summarizing Retrievals using Answer Candidates for Open-domain QA of LLMs
https://openreview.net/forum?id=w4DW6qkRmt
[ "Jaehyung Kim", "Jaehyun Nam", "Sangwoo Mo", "Jongjin Park", "Sang-Woo Lee", "Minjoon Seo", "Jung-Woo Ha", "Jinwoo Shin" ]
Poster
generative models
Large language models (LLMs) have made significant advancements in various natural language processing tasks, including question answering (QA) tasks. While incorporating new information with the retrieval of relevant passages is a promising way to improve QA with LLMs, the existing methods often require additional fin...
[ "question answering", "large language model", "retrieval" ]
We propose a simple framework to improve ODQA accuracy of LLM, by generating conditional summarizations of retrieval and evaluating them with carefully desinged prompts.
8,921
2404.13081
title_snapshot
[ -0.03135789558291435, -0.04272252321243286, 0.009311421774327755, 0.044135384261608124, 0.05928784981369972, 0.00569961778819561, 0.007755275350064039, 0.013205264694988728, -0.028915930539369583, 0.015328964218497276, -0.05431986227631569, 0.04508646950125694, -0.05441197007894516, -0.023...
Retrieval meets Long Context Large Language Models
https://openreview.net/forum?id=xw5nxFWMlo
[ "Peng Xu", "Wei Ping", "Xianchao Wu", "Lawrence McAfee", "Chen Zhu", "Zihan Liu", "Sandeep Subramanian", "Evelina Bakhturina", "Mohammad Shoeybi", "Bryan Catanzaro" ]
Poster
generative models
Extending the context window of large language models (LLMs) is getting popular recently, while the solution of augmenting LLMs with retrieval has existed for years. The natural questions are: i) Retrieval-augmentation versus long context window, which one is better for downstream tasks? ii) Can both methods be combine...
[ "Large Language Models", "Long Context Window", "Retrieval" ]
null
8,917
2310.03025
title_snapshot
[ -0.019022872671484947, -0.026305340230464935, -0.002397110452875495, 0.019671007990837097, 0.038405947387218475, -0.016836801543831825, 0.013870264403522015, 0.04597173631191254, -0.02646830305457115, -0.010850939899682999, -0.04920180141925812, 0.041513893753290176, -0.04443778842687607, ...
Neural Spectral Methods: Self-supervised learning in the spectral domain
https://openreview.net/forum?id=2DbVeuoa6a
[ "Yiheng Du", "Nithin Chalapathi", "Aditi S. Krishnapriyan" ]
Poster
applications to physical sciences (physics, chemistry, biology, etc.)
We present Neural Spectral Methods, a technique to solve parametric Partial Differential Equations (PDEs), grounded in classical spectral methods. Our method uses orthogonal bases to learn PDE solutions as mappings between spectral coefficients, instantiating a spectral-based neural operator. In contrast to current mac...
[ "Machine learning for PDEs", "spectral methods", "neural network differentiation", "spectral loss", "PDEs", "neural operators" ]
We present Neural Spectral Methods to solve parametric PDEs in the spectral domain.
8,910
2312.05225
title_snapshot
[ -0.05426158010959625, -0.020853808149695396, 0.008294221945106983, 0.04428879916667938, 0.01422253716737032, 0.035701919347047806, 0.030363645404577255, -0.03967275843024254, -0.012166899628937244, -0.07311785966157913, 0.01166885532438755, 0.006299881264567375, -0.053775280714035034, 0.02...
Kosmos-G: Generating Images in Context with Multimodal Large Language Models
https://openreview.net/forum?id=he6mX9LTyE
[ "Xichen Pan", "Li Dong", "Shaohan Huang", "Zhiliang Peng", "Wenhu Chen", "Furu Wei" ]
Poster
generative models
Recent advancements in subject-driven image generation have made significant strides. However, current methods still fall short in diverse application scenarios, as they require test-time tuning and cannot accept interleaved multi-image and text input. These limitations keep them far from the ultimate goal of "image as...
[ "Diffusion Models", "Vision-Language", "Multimodal Large Language Model", "Image Generation", "Subject-Driven Generation" ]
Kosmos-G leverages Multimodal Large Language Models for subject-driven image generation
8,897
2310.02992
title_snapshot
[ 0.015893783420324326, -0.009419366717338562, 0.011497197672724724, 0.03811556100845337, 0.030839236453175545, 0.017515523359179497, 0.020913785323500633, 0.04756830632686615, -0.04143744707107544, -0.0159671138972044, -0.04305757209658623, 0.021929971873760223, -0.08176266402006149, -0.000...
Selective Visual Representations Improve Convergence and Generalization for Embodied AI
https://openreview.net/forum?id=kC5nZDU5zf
[ "Ainaz Eftekhar", "Kuo-Hao Zeng", "Jiafei Duan", "Ali Farhadi", "Aniruddha Kembhavi", "Ranjay Krishna" ]
Spotlight
reinforcement learning
Embodied AI models often employ off the shelf vision backbones like CLIP to encode their visual observations. Although such general purpose representations encode rich syntactic and semantic information about the scene, much of this information is often irrelevant to the specific task at hand. This introduces noise wit...
[ "Embodied-AI", "Task-conditioned Representations", "Visual Navigation", "Reinforcement Learning" ]
null
8,895
2311.04193
title_snapshot
[ 0.029402313753962517, 0.012562412768602371, -0.0033254469744861126, 0.016010098159313202, 0.019963877275586128, 0.031127087771892548, 0.031301289796829224, 0.022391008213162422, -0.04765595495700836, -0.047123637050390244, -0.047365788370370865, -0.007736259140074253, -0.09575794637203217, ...
Assessing Uncertainty in Similarity Scoring: Performance & Fairness in Face Recognition
https://openreview.net/forum?id=lAhQCHuANV
[ "Jean-Rémy Conti", "Stephan Clémençon" ]
Poster
societal considerations including fairness, safety, privacy
The ROC curve is the major tool for assessing not only the performance but also the fairness properties of a similarity scoring function. In order to draw reliable conclusions based on empirical ROC analysis, accurately evaluating the uncertainty level related to statistical versions of the ROC curves of interest is ab...
[ "Uncertainty", "Face", "Recognition", "Performance", "ROC", "Fairness", "Bootstrap" ]
This paper provides a valid bootstrap method for quantifying the uncertainty of ROC-based metrics of a similarity scoring function, such as performance and fairness metrics in Face Recognition.
8,889
2211.07245
title_snapshot
[ 0.0007845181971788406, -0.0031747580505907536, -0.0136029701679945, 0.01631234772503376, 0.03558370843529701, 0.03656204044818878, 0.017403343692421913, 0.005304731894284487, -0.007291954476386309, -0.051812607795000076, -0.030173655599355698, 0.0005089329788461328, -0.07955354452133179, -...
LitCab: Lightweight Language Model Calibration over Short- and Long-form Responses
https://openreview.net/forum?id=jH67LHVOIO
[ "Xin Liu", "Muhammad Khalifa", "Lu Wang" ]
Poster
generative models
A model is considered well-calibrated when its probability estimate aligns with the actual likelihood of the output being correct. Calibrating language models (LMs) is crucial, as it plays a vital role in detecting and mitigating hallucinations of LMs as well as building more trustworthy models. However, standard calib...
[ "calibration", "hallucination", "large language model" ]
null
8,885
2310.19208
title_snapshot
[ -0.005374427884817123, -0.007025012746453285, 0.007172692101448774, 0.021077821031212807, 0.027566375210881233, 0.025652598589658737, 0.03102581575512886, 0.03927149996161461, -0.02022576332092285, -0.014172476716339588, -0.014928275719285011, 0.034509941935539246, -0.05243639647960663, -0...
Improving Generalization of Alignment with Human Preferences through Group Invariant Learning
https://openreview.net/forum?id=fwCoLe3TAX
[ "Rui Zheng", "Wei Shen", "Yuan Hua", "Wenbin Lai", "Shihan Dou", "Yuhao Zhou", "Zhiheng Xi", "Xiao Wang", "Haoran Huang", "Tao Gui", "Qi Zhang", "Xuanjing Huang" ]
Spotlight
representation learning for computer vision, audio, language, and other modalities
The success of AI assistants based on language models (LLMs) hinges crucially on Reinforcement Learning from Human Feedback (RLHF), which enables the generation of responses more aligned with human preferences. As universal AI assistants, there's a growing expectation for them to perform consistently across various do...
[ "alignment", "language model", "invariant learning" ]
This paper introduces a novel method for aligning AI assistants with human preferences, boosting RLHF training stability and improving the model’s generalization across various domains.
8,880
2310.11971
title_snapshot
[ 0.004047398455440998, -0.010331218130886555, 0.02158411405980587, 0.036589063704013824, 0.028925566002726555, 0.019178051501512527, 0.03288806974887848, 0.02098918706178665, -0.012378162704408169, -0.026264047250151634, -0.05497309938073158, 0.046755101531744, -0.09860115498304367, -0.0389...
Chain-of-Knowledge: Grounding Large Language Models via Dynamic Knowledge Adapting over Heterogeneous Sources
https://openreview.net/forum?id=cPgh4gWZlz
[ "Xingxuan Li", "Ruochen Zhao", "Yew Ken Chia", "Bosheng Ding", "Shafiq Joty", "Soujanya Poria", "Lidong Bing" ]
Poster
representation learning for computer vision, audio, language, and other modalities
We present chain-of-knowledge (CoK), a novel framework that augments large language models (LLMs) by dynamically incorporating grounding information from heterogeneous sources. It results in more factual rationales and reduced hallucination in generation. Specifically, CoK consists of three stages: reasoning preparati...
[ "large language model", "knowledge grounding" ]
We present chain-of-knowledge, a novel framework that augments large language models dynamically by incorporating grounding information from heterogeneous sources.
8,874
2305.13269
title_snapshot
[ -0.014428305439651012, -0.023457970470190048, -0.017709342762827873, 0.03522725775837898, 0.03903990611433983, 0.005040234420448542, 0.025753425434231758, -0.004019958432763815, -0.02335711009800434, 0.027301445603370667, -0.027725283056497574, 0.05301584675908089, -0.0659012496471405, -0....
Policy Rehearsing: Training Generalizable Policies for Reinforcement Learning
https://openreview.net/forum?id=m3xVPaZp6Z
[ "Chengxing Jia", "Chenxiao Gao", "Hao Yin", "Fuxiang Zhang", "Xiong-Hui Chen", "Tian Xu", "Lei Yuan", "Zongzhang Zhang", "Zhi-Hua Zhou", "Yang Yu" ]
Poster
reinforcement learning
Human beings can make adaptive decisions in a preparatory manner, i.e., by making preparations in advance, which offers significant advantages in scenarios where both online and offline experiences are expensive and limited. Meanwhile, current reinforcement learning methods commonly rely on numerous environment interac...
[ "Reinforcement Learning", "Model-based Reinforcement Learning", "Offline Reinforcement Learning" ]
We introduce the idea of rehearsal into reinforcement learning in scenarios where both online and offline experiences are expensive and limited.
8,871
null
null
[ -0.02276040054857731, -0.032548218965530396, -0.007771264761686325, 0.057694364339113235, 0.05205531790852547, 0.030296603217720985, 0.02972353808581829, 0.0016151823801919818, -0.03935481607913971, -0.031603336334228516, -0.034560736268758774, 0.04106862470507622, -0.08583425730466843, -0...
Energy-based Automated Model Evaluation
https://openreview.net/forum?id=CHGcP6lVWd
[ "Ru Peng", "Heming Zou", "Haobo Wang", "Yawen Zeng", "Zenan Huang", "Junbo Zhao" ]
Poster
representation learning for computer vision, audio, language, and other modalities
The conventional evaluation protocols on machine learning models rely heavily on a labeled, i.i.d-assumed testing dataset, which is not often present in real-world applications. The Automated Model Evaluation (AutoEval) shows an alternative to this traditional workflow, by forming a proximal prediction pipeline of the ...
[ "Automated Model Evalutaion", "Energy", "Meta-distribution", "Distribution shift" ]
We introduce a simple yet effective method (Meta-Distribution Energy) in predicting a model's generalization to previously unseen, unlabeled datasets with theoretical guarantees and achieves state-of-the-art performance.
8,869
2401.12689
title_snapshot
[ -0.01222179178148508, -0.01716836728155613, -0.006489584222435951, 0.05342083051800728, 0.045485977083444595, 0.011165712028741837, 0.032643597573041916, -0.004489208571612835, -0.03468150272965431, -0.03769489750266075, 0.019501740112900734, 0.01640939526259899, -0.0830496996641159, -0.00...
Deceptive Fairness Attacks on Graphs via Meta Learning
https://openreview.net/forum?id=iS5ADHNg2A
[ "Jian Kang", "Yinglong Xia", "Ross Maciejewski", "Jiebo Luo", "Hanghang Tong" ]
Poster
learning on graphs and other geometries & topologies
We study deceptive fairness attacks on graphs to answer the following question: How can we achieve poisoning attacks on a graph learning model to exacerbate the bias deceptively? We answer this question via a bi-level optimization problem and propose a meta learning-based framework named FATE. FATE is broadly applicabl...
[ "graph learning", "fairness", "adversarial attacks" ]
We develop a meta learning-based poisoning attack strategy to exacerbate unfairness of graph learning models, while preserving the utility in downstream tasks.
8,867
2310.15653
title_snapshot
[ -0.012839334085583687, -0.012867653742432594, -0.012039507739245892, 0.052427131682634354, 0.011763589456677437, 0.006528205703943968, 0.0257936529815197, 0.009756017476320267, -0.026497721672058105, -0.0247302558273077, 0.0286736898124218, -0.00784658920019865, -0.06931185722351074, 0.013...
What Matters to You? Towards Visual Representation Alignment for Robot Learning
https://openreview.net/forum?id=CTlUHIKF71
[ "Thomas Tian", "Chenfeng Xu", "Masayoshi Tomizuka", "Jitendra Malik", "Andrea Bajcsy" ]
Poster
applications to robotics, autonomy, planning
When operating in service of people, robots need to optimize rewards aligned with end-user preferences. Since robots will rely on raw perceptual inputs, their rewards will inevitably use visual representations. Recently there has been excitement in using representations from pre-trained visual models, but key to making...
[ "Robot learning", "Preference learning", "Visual reward learning", "Representation alignment" ]
null
8,866
2310.07932
title_snapshot
[ 0.003927988465875387, 0.016830412670969963, -0.012310435995459557, 0.03616226837038994, 0.017289895564317703, 0.021854272112250328, 0.010168804787099361, 0.03928424417972565, -0.06394922733306885, -0.0453936830163002, -0.068433977663517, 0.014435895718634129, -0.08752382546663284, -0.01234...
PINNACLE: PINN Adaptive ColLocation and Experimental points selection
https://openreview.net/forum?id=GzNaCp6Vcg
[ "Gregory Kang Ruey Lau", "Apivich Hemachandra", "See-Kiong Ng", "Bryan Kian Hsiang Low" ]
Spotlight
neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)
Physics-Informed Neural Networks (PINNs), which incorporate PDEs as soft constraints, train with a composite loss function that contains multiple training point types: different types of collocation points chosen during training to enforce each PDE and initial/boundary conditions, and experimental points which are usua...
[ "Physics-informed Neural Networks", "PINNs", "adaptive training points selection" ]
A novel PINN training algorithm, motivated by analysis of the Neural Tangent Kernel, that jointly selects all training point types in the composite loss function to gain large performance boosts for forward, inverse, and transfer learning problems.
8,858
2404.07662
title_snapshot
[ -0.05723930895328522, 0.010514226742088795, 0.001656428794376552, 0.03084619902074337, 0.04151275008916855, 0.02319612354040146, -0.0026173503138124943, -0.051839787513017654, -0.06763726472854614, -0.0293080136179924, -0.018543019890785217, 0.007471909746527672, -0.03683885931968689, 0.02...
FedDA: Faster Adaptive Gradient Methods for Federated Constrained Optimization
https://openreview.net/forum?id=kjn99xFUF3
[ "Junyi Li", "Feihu Huang", "Heng Huang" ]
Poster
general machine learning (i.e., none of the above)
Federated learning (FL) is an emerging learning paradigm where a set of distributed clients learns a task under the coordination of a server. The FedAvg algorithm is one of the most widely used methods in FL. In FedAvg, the learning rate is a constant rather than changing adaptively. Adaptive gradient methods have demo...
[ "Federated Learning", "Adaptive Gradient Methods" ]
null
8,857
null
null
[ -0.01552658248692751, -0.05055617168545723, 0.007007535547018051, 0.024128712713718414, 0.04579473286867142, 0.03801934793591499, 0.027789220213890076, -0.0025084151420742273, -0.01795285940170288, -0.03801528736948967, 0.005823610816150904, 0.0025866376236081123, -0.05592818930745125, 0.0...
BooookScore: A systematic exploration of book-length summarization in the era of LLMs
https://openreview.net/forum?id=7Ttk3RzDeu
[ "Yapei Chang", "Kyle Lo", "Tanya Goyal", "Mohit Iyyer" ]
Oral
datasets and benchmarks
Summarizing book-length documents ($>$100K tokens) that exceed the context window size of large language models (LLMs) requires first breaking the input document into smaller chunks and then prompting an LLM to merge, update, and compress chunk-level summaries. Despite the complexity and importance of this task, it ha...
[ "summarization", "evaluation", "long context", "prompting", "LLM" ]
null
8,848
2310.00785
title_snapshot
[ -0.038996390998363495, -0.04441021382808685, -0.01211413461714983, -0.00010627735173329711, 0.027865873649716377, -0.02332059107720852, 0.012192034162580967, 0.025608964264392853, -0.020611397922039032, -0.0016761802835389972, -0.04096684977412224, 0.025920286774635315, -0.056855253875255585...
Extending Power of Nature from Binary to Real-Valued Graph Learning in Real World
https://openreview.net/forum?id=qT7DXUmX7j
[ "Chunshu Wu", "Ruibing Song", "Chuan Liu", "Yunan Yang", "Ang Li", "Michael Huang", "Tong Geng" ]
Poster
learning on graphs and other geometries & topologies
Nature performs complex computations constantly at clearly lower cost and higher performance than digital computers. It is crucial to understand how to harness the unique computational power of nature in Machine Learning (ML). In the past decade, besides the development of Neural Networks (NNs), the community has also ...
[ "graph learning", "nature-powered computing", "dynamic physical system" ]
We upgrade a binary Ising Machine and its associated model to support real values in solving real-world problems, achieving orders of magnitude of speedup and energy efficiency in Graph Learning compared to baseline GNNs
8,845
null
null
[ -0.023815544322133064, -0.04954397305846214, -0.006649492308497429, 0.046107713133096695, 0.04962506890296936, 0.016465870663523674, 0.03413298726081848, 0.025718756020069122, -0.018013736233115196, -0.038885071873664856, 0.009104697965085506, -0.0084299948066473, -0.07396344840526581, 0.0...
Meta-VBO: Utilizing Prior Tasks in Optimizing Risk Measures with Gaussian Processes
https://openreview.net/forum?id=ElykcDu5YK
[ "Quoc Phong Nguyen", "Bryan Kian Hsiang Low", "Patrick Jaillet" ]
Poster
probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)
Research on optimizing the risk measure of a blackbox function using Gaussian processes, especially Bayesian optimization (BO) of risk measures, has become increasingly important due to the inevitable presence of uncontrollable variables in real-world applications. Nevertheless, existing works on BO of risk measures st...
[ "meta-learning", "Bayesian optimization", "risk measure", "value-at-risk", "conditional value-at-risk" ]
null
8,829
null
null
[ 0.006872519850730896, 0.008498817682266235, 0.003645704360678792, 0.01904851570725441, 0.03748917952179909, 0.04935915768146515, 0.03767627105116844, -0.00839681550860405, -0.020586973056197166, -0.03824751451611519, -0.01515047438442707, 0.04194340109825134, -0.07509715110063553, -0.02802...
Data Debugging with Shapley Importance over Machine Learning Pipelines
https://openreview.net/forum?id=qxGXjWxabq
[ "Bojan Karlaš", "David Dao", "Matteo Interlandi", "Sebastian Schelter", "Wentao Wu", "Ce Zhang" ]
Poster
infrastructure, software libraries, hardware, etc.
When a machine learning (ML) model exhibits poor quality (e.g., poor accuracy or fairness), the problem can often be traced back to errors in the training data. Being able to discover the data examples that are the most likely culprits is a fundamental concern that has received a lot of attention recently. One prominen...
[ "data debugging", "data valuation", "shapley value", "machine learning pipelines" ]
Efficiently computing the Shapley value of training data examples over machine learning pipelines.
8,826
2204.11131
title_judge
[ -0.026307815685868263, -0.028472712263464928, -0.03329808637499809, 0.057797808200120926, 0.03979548439383507, 0.02300920896232128, 0.030848756432533264, -0.024826066568493843, -0.020907549187541008, -0.04484668746590614, -0.024255795404314995, 0.00855664536356926, -0.07822078466415405, 0....
Learning to Relax: Setting Solver Parameters Across a Sequence of Linear System Instances
https://openreview.net/forum?id=5t57omGVMw
[ "Mikhail Khodak", "Edmond Chow", "Maria Florina Balcan", "Ameet Talwalkar" ]
Spotlight
learning theory
Solving a linear system ${\bf Ax}={\bf b}$ is a fundamental scientific computing primitive for which numerous solvers and preconditioners have been developed. These come with parameters whose optimal values depend on the system being solved and are often impossible or too expensive to identify; thus in practice sub-...
[ "scientific computing", "data-driven algorithm design", "online learning", "multi-armed bandits", "contextual bandits", "numerical analysis", "learning-augmented algorithms", "algorithms with predictions" ]
We show provable guarantees for learning the relaxation parameter of linear system solvers.
8,824
2310.02246
title_snapshot
[ -0.06398044526576996, -0.003976070787757635, 0.004344920627772808, 0.05718763545155525, 0.02532850205898285, 0.03446374833583832, 0.021425781771540642, -0.005685683339834213, -0.005787413101643324, -0.04735869914293289, -0.017053501680493355, -0.011869841255247593, -0.08321982622146606, -0...
Pessimistic Nonlinear Least-Squares Value Iteration for Offline Reinforcement Learning
https://openreview.net/forum?id=4kLVvIh8cp
[ "Qiwei Di", "Heyang Zhao", "Jiafan He", "Quanquan Gu" ]
Poster
reinforcement learning
Offline reinforcement learning (RL), where the agent aims to learn the optimal policy based on the data collected by a behavior policy, has attracted increasing attention in recent years. While offline RL with linear function approximation has been extensively studied with optimal results achieved under certain assumpt...
[ "Offline reinforcement learning", "instance-dependent", "least-squares value iteration" ]
In this paper, we present Pessimistic Nonlinear Least-Square Value Iteration (PNLSVI), an oracle-efficient algorithm for offline RL with non-linear function approximation.
8,813
2310.01380
title_snapshot
[ -0.03958132490515709, -0.012224400416016579, 0.004282488487660885, 0.044506508857011795, 0.03130444511771202, 0.05489762872457504, 0.007556870114058256, -0.005293384660035372, -0.04504607990384102, -0.024810628965497017, -0.042848438024520874, 0.018765205517411232, -0.0811479240655899, -0....
SKILL-MIX: a Flexible and Expandable Family of Evaluations for AI Models
https://openreview.net/forum?id=Jf5gplvglq
[ "Dingli Yu", "Simran Kaur", "Arushi Gupta", "Jonah Brown-Cohen", "Anirudh Goyal", "Sanjeev Arora" ]
Poster
datasets and benchmarks
With LLMs shifting their role from statistical modeling of language to serving as general-purpose AI agents, how should LLM evaluations change? Arguably, a key ability of an AI agent is to flexibly combine, as needed, the basic skills it has learned. The capability to combine skills plays an important role in (human) p...
[ "Large language model", "skill evaluation", "LLM benchmark", "emergence" ]
null
8,809
2310.17567
title_snapshot
[ 0.0008029423770494759, -0.01532126683741808, -0.031675588339567184, 0.017675597220659256, 0.051389098167419434, 0.016153153032064438, 0.02298312447965145, 0.007815886288881302, -0.02402820996940136, -0.023805487900972366, -0.028415534645318985, 0.04325418174266815, -0.06583692133426666, -0...
A Quadratic Synchronization Rule for Distributed Deep Learning
https://openreview.net/forum?id=yroyhkhWS6
[ "Xinran Gu", "Kaifeng Lyu", "Sanjeev Arora", "Jingzhao Zhang", "Longbo Huang" ]
Poster
optimization
In distributed deep learning with data parallelism, synchronizing gradients at each training step can cause a huge communication overhead, especially when many nodes work together to train large models. Local gradient methods, such as Local SGD, address this issue by allowing workers to compute locally for $H$ steps ...
[ "distributed training", "Local SGD", "local gradient methods", "generalization", "implicit bias", "sharpness" ]
We propose Quadratic Synchronization Rule (QSR) to dynamically set the synchronization period in local gradient methods based on the learning rate, reducing wall-clock training time and improving test accuracy of ResNet and ViT on ImageNet.
8,805
2310.14423
title_snapshot
[ -0.015337939374148846, -0.024808157235383987, -0.006555766332894564, 0.03407871723175049, 0.017828207463026047, 0.027547454461455345, 0.028842918574810028, 0.0016025339718908072, -0.041164472699165344, -0.058617182075977325, -0.006886619143188, -0.022987263277173042, -0.07048231363296509, ...
ArchLock: Locking DNN Transferability at the Architecture Level with a Zero-Cost Binary Predictor
https://openreview.net/forum?id=e2YOVTenU9
[ "Tong Zhou", "Shaolei Ren", "Xiaolin Xu" ]
Poster
societal considerations including fairness, safety, privacy
Deep neural network (DNN) models, despite their impressive performance, are vulnerable to exploitation by attackers who attempt to transfer them to other tasks for their own benefit. Current defense strategies mainly address this vulnerability at the model parameter level, leaving the potential of architectural-level d...
[ "Defense; DNN Transferability; Neural Architecture Search" ]
null
8,804
null
null
[ -0.027991853654384613, 0.002420809119939804, -0.017667794600129128, 0.03236870467662811, 0.06253501027822495, 0.040665000677108765, 0.0324985608458519, -0.013173501007258892, -0.022970158606767654, -0.029944896697998047, 0.020659182220697403, -0.02509709633886814, -0.04959944263100624, 0.0...
Leftover Lunch: Advantage-based Offline Reinforcement Learning for Language Models
https://openreview.net/forum?id=ZDGKPbF0VQ
[ "Ashutosh Baheti", "Ximing Lu", "Faeze Brahman", "Ronan Le Bras", "Maarten Sap", "Mark Riedl" ]
Poster
reinforcement learning
Reinforcement Learning with Human Feedback (RLHF) is the most prominent method for Language Model (LM) alignment. However, RLHF is an unstable and data-hungry process that continually requires new high-quality LM-generated data for finetuning. We introduce Advantage-Leftover Lunch RL (A-LoL), a new class of offline pol...
[ "Reinforcement Learning", "Natural Language Generation", "Offline Policy Gradients" ]
Advantage Leftover Lunch RL (A-LoL), a simple training algorithm that uses offline policy gradients for learning language generation tasks as a single action RL game.
8,769
2305.14718
title_snapshot
[ -0.020134711638092995, -0.010629581287503242, -0.002270705532282591, 0.04468616843223572, 0.03683939203619957, 0.02671089395880699, 0.006691685412079096, 0.029270347207784653, -0.03991779685020447, -0.009499943815171719, -0.03237584978342056, 0.057674311101436615, -0.08856743574142456, -0....
RECOMP: Improving Retrieval-Augmented LMs with Context Compression and Selective Augmentation
https://openreview.net/forum?id=mlJLVigNHp
[ "Fangyuan Xu", "Weijia Shi", "Eunsol Choi" ]
Poster
representation learning for computer vision, audio, language, and other modalities
Retrieval-augmented language models improve language models (LMs) by retrieving documents and prepending them in-context. However, these documents, often spanning hundreds of words, make inference substantially less efficient. We propose compressing the retrieved documents into textual summaries prior to in-context int...
[ "retrieval augmented language model", "language modeling", "question answering", "summarization", "distillation" ]
We train compressor models to shorten retrieved documents passed to language models.
8,767
2310.04408
title_judge
[ -0.018926313146948814, -0.016586117446422577, -0.022822391241788864, 0.04228972643613815, 0.0542873740196228, 0.01713266782462597, 0.019051168113946915, 0.029686346650123596, -0.025326766073703766, -0.014947672374546528, -0.043277349323034286, 0.04412791505455971, -0.05722393840551376, -0....
Gen-Z: Generative Zero-Shot Text Classification with Contextualized Label Descriptions
https://openreview.net/forum?id=rkplYfqUr0
[ "Sachin Kumar", "Chan Young Park", "Yulia Tsvetkov" ]
Poster
representation learning for computer vision, audio, language, and other modalities
Language model (LM) prompting—a popular paradigm for solving NLP tasks—has been shown to be susceptible to miscalibration and brittleness to slight prompt variations, caused by its discriminative prompting approach, i.e., predicting the label given the input. To address these issues, we propose Gen-Z—a generative promp...
[ "zero-shot classification", "prompting", "generative classification", "label descriptions" ]
We present a zero-shot prompting framework that outperforms or closely matches in-context learning for a variety of text classification tasks.
8,763
2311.07115
title_snapshot
[ 0.010458094999194145, -0.02110271528363228, -0.028644494712352753, 0.0631520226597786, 0.041530560702085495, 0.03141271695494652, 0.005185327026993036, 0.029879925772547722, 0.016966745257377625, 0.007219075690954924, -0.05293450504541397, 0.04209389165043831, -0.05809210240840912, 0.02285...
In-Context Learning Dynamics with Random Binary Sequences
https://openreview.net/forum?id=62K7mALO2q
[ "Eric J Bigelow", "Ekdeep Singh Lubana", "Robert P. Dick", "Hidenori Tanaka", "Tomer Ullman" ]
Poster
visualization or interpretation of learned representations
Large language models (LLMs) trained on huge text datasets demonstrate intriguing capabilities, achieving state-of-the-art performance on tasks they were not explicitly trained for. The precise nature of LLM capabilities is often mysterious, and different prompts can elicit different capabilities through in-context lea...
[ "In-Context Learning", "Large Language Models", "Interpretability", "Computational Cognitive Science" ]
null
8,758
2310.17639
title_snapshot
[ -0.015117088332772255, -0.0005940566188655794, -0.03312806040048599, 0.05309561640024185, 0.02746463380753994, 0.012533753179013729, 0.026570072397589684, 0.05545322969555855, -0.047433435916900635, 0.00658431276679039, -0.0329340361058712, 0.02244800142943859, -0.031197868287563324, -0.01...