pair_id int64 0 99 | insight stringlengths 650 2.74k | paper1_scores float64 -5,484.72 -835.95 | paper2_scores float64 -6,548.31 -950.63 | joint_scores float64 -5,539.25 -981.31 | no_context_scores float64 -6,717.06 -578.01 | contrastive_loss float64 1.38k 14k | paper1_scores_avg float64 -15.41 -7.71 | paper2_scores_avg float64 -15.35 -7.96 | joint_scores_avg float64 -15.56 -7.5 | no_context_scores_avg float64 -15.55 -5.78 | abstract listlengths 2 2 | valid_abstracts bool 1
class |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
0 | <insight>
Advanced AI systems often need to operate within complex behavioral boundaries, such as adhering to safety protocols or aligning with nuanced human preferences. One approach involves learning these boundaries implicitly as latent structures from available data, allowing policies to optimize objectives while r... | -1,362.5 | -2,230.5625 | -1,856.304688 | -2,224.21875 | 3,960.976563 | -7.921512 | -12.968387 | -10.792469 | -12.931504 | [
"LATENT SAFETY-CONSTRAINED POLICY APPROACH\nFOR SAFE OFFLINE REINFORCEMENT LEARNING\n\nPrajwal Koirala, Zhanhong Jiang, Soumik Sarkar & Cody Fleming\nIowa State University\nAmes, Iowa, USA\n{prajwal,zhjiang,soumiks,flemingc}@iastate.edu\n\nABSTRACT\n\nIn safe offline reinforcement learning (RL), the objective is to... | true |
0 | <insight>
Effectively guiding intelligent systems requires learning complex objectives that often involve balancing desired outcomes with critical constraints, such as adhering to safety protocols or aligning with nuanced human values. Two distinct strategies are emerging to tackle this challenge. One approach involves... | -1,440.859375 | -2,338.046875 | -1,824.90625 | -2,236.835938 | 4,190.835938 | -8.426078 | -13.672789 | -10.671967 | -13.080912 | [
"LATENT SAFETY-CONSTRAINED POLICY APPROACH\nFOR SAFE OFFLINE REINFORCEMENT LEARNING\n\nPrajwal Koirala, Zhanhong Jiang, Soumik Sarkar & Cody Fleming\nIowa State University\nAmes, Iowa, USA\n{prajwal,zhjiang,soumiks,flemingc}@iastate.edu\n\nABSTRACT\n\nIn safe offline reinforcement learning (RL), the objective is to... | true |
0 | <insight>Guiding complex agent behavior in reinforcement learning often requires more than just maximizing a predefined reward. Two distinct but complementary strategies emerge for learning these crucial guiding signals, especially when direct supervision or comprehensive data is lacking. One strategy focuses on inferr... | -1,409.46875 | -1,568.875 | -1,117.25 | -1,579.492188 | 3,440.585938 | -9.787977 | -10.894965 | -7.75868 | -10.968696 | [
"LATENT SAFETY-CONSTRAINED POLICY APPROACH\nFOR SAFE OFFLINE REINFORCEMENT LEARNING\n\nPrajwal Koirala, Zhanhong Jiang, Soumik Sarkar & Cody Fleming\nIowa State University\nAmes, Iowa, USA\n{prajwal,zhjiang,soumiks,flemingc}@iastate.edu\n\nABSTRACT\n\nIn safe offline reinforcement learning (RL), the objective is to... | true |
1 | <insight>
Achieving effective simultaneous alignment of representations from multiple distinct sources (like vision, audio, and text) within a shared high-dimensional embedding space requires addressing challenges beyond just the alignment mechanism itself. While geometric approaches, such as minimizing the volume span... | -1,300.796875 | -1,941.8125 | -1,987.75 | -1,831.402344 | 3,086.261719 | -8.446733 | -12.609172 | -12.907468 | -11.892223 | [
"GRAMIAN MULTIMODAL REPRESENTATION\nLEARNING AND ALIGNMENT\n\nGiordano Cicchetti∗, Eleonora Grassucci∗, Luigi Sigillo, Danilo Comminiello\nDept. of Information Engineering, Electronics, and Telecomm., Sapienza University of Rome, Italy\n{name.surname}@uniroma1.it\n\nABSTRACT\n\nHuman perception integrates multiple ... | true |
1 | <insight>
Developing advanced machine learning systems that operate effectively in high-dimensional latent or embedding spaces necessitates addressing two distinct but complementary challenges. Firstly, when integrating information from multiple sources (like different senses or data types), ensuring that their respect... | -1,865.929688 | -2,542.195313 | -2,586.5625 | -2,524.289063 | 4,345.851563 | -10.140923 | -13.816278 | -14.057405 | -13.718963 | [
"GRAMIAN MULTIMODAL REPRESENTATION\nLEARNING AND ALIGNMENT\n\nGiordano Cicchetti∗, Eleonora Grassucci∗, Luigi Sigillo, Danilo Comminiello\nDept. of Information Engineering, Electronics, and Telecomm., Sapienza University of Rome, Italy\n{name.surname}@uniroma1.it\n\nABSTRACT\n\nHuman perception integrates multiple ... | true |
1 | <insight>
Advancements are simultaneously being made in how multiple distinct information streams (like vision, audio, and text) are integrated and how generative models operate within abstract representational spaces. One line of work focuses on ensuring that the vector representations derived from different sources a... | -2,318.25 | -2,876.84375 | -2,973.78125 | -2,852.25 | 5,073.5625 | -11.199275 | -13.897796 | -14.366093 | -13.778986 | [
"GRAMIAN MULTIMODAL REPRESENTATION\nLEARNING AND ALIGNMENT\n\nGiordano Cicchetti∗, Eleonora Grassucci∗, Luigi Sigillo, Danilo Comminiello\nDept. of Information Engineering, Electronics, and Telecomm., Sapienza University of Rome, Italy\n{name.surname}@uniroma1.it\n\nABSTRACT\n\nHuman perception integrates multiple ... | true |
4 | <insight>While foundational theoretical work establishes guarantees for how learning algorithms can converge to optimal policies, even in complex, long-term scenarios, the practical assessment of sophisticated AI systems presents its own distinct set of challenges. Specifically, verifying the true capabilities of syste... | -1,819.6875 | -1,999.492188 | -2,005.96875 | -1,771.75 | 3,584.960938 | -12.378826 | -13.601988 | -13.646046 | -12.052721 | [
"GLOBAL CONVERGENCE OF POLICY GRADIENT IN AV-\nERAGE REWARD MDPS\n\nNavdeep Kumar∗\nElectrical and Computer Engineering\nTechnion - Israel Institute of Technology\nnavdeepkumar@campus.technion.ac.il\n\nYashaswini Murthy∗\nECE & CSL\nUniversity of Illinois Urbana-Champaign\nymurthy2@illinois.edu\n\nItai Shufaro\nEle... | true |
4 | <insight>
Significant effort is dedicated to rigorously proving the theoretical convergence and performance guarantees of complex learning algorithms, ensuring they can, in principle, reach optimal solutions even in challenging settings like infinite-horizon average-reward problems. However, the practical evaluation of... | -1,742.203125 | -1,939.3125 | -1,947.25 | -1,528.787109 | 3,263.052734 | -12.015194 | -13.374569 | -13.429311 | -10.54336 | [
"GLOBAL CONVERGENCE OF POLICY GRADIENT IN AV-\nERAGE REWARD MDPS\n\nNavdeep Kumar∗\nElectrical and Computer Engineering\nTechnion - Israel Institute of Technology\nnavdeepkumar@campus.technion.ac.il\n\nYashaswini Murthy∗\nECE & CSL\nUniversity of Illinois Urbana-Champaign\nymurthy2@illinois.edu\n\nItai Shufaro\nEle... | true |
4 | <insight>
Advancing artificial intelligence involves tackling complexity on two complementary fronts. On one hand, establishing foundational understanding requires rigorous theoretical analysis of learning algorithms, particularly under challenging, long-term scenarios (like optimizing for average rewards over infinite... | -2,230.78125 | -2,445.34375 | -2,432.21875 | -2,158.222656 | 4,402.128906 | -13.199889 | -14.46949 | -14.391827 | -12.770548 | [
"GLOBAL CONVERGENCE OF POLICY GRADIENT IN AV-\nERAGE REWARD MDPS\n\nNavdeep Kumar∗\nElectrical and Computer Engineering\nTechnion - Israel Institute of Technology\nnavdeepkumar@campus.technion.ac.il\n\nYashaswini Murthy∗\nECE & CSL\nUniversity of Illinois Urbana-Champaign\nymurthy2@illinois.edu\n\nItai Shufaro\nEle... | true |
5 | <insight>
Optimizing neural networks for specific goals, such as maximizing efficiency through compression or enhancing capability via large-scale pretraining for downstream tasks, reveals a common challenge: naive application of seemingly beneficial techniques can unexpectedly degrade the ultimate desired performance.... | -2,102.28125 | -2,250.40625 | -1,994.140625 | -2,299.359375 | 4,657.90625 | -12.013036 | -12.859465 | -11.395089 | -13.139196 | [
"QP-SNNS: QUANTIZED AND PRUNED SPIKING\nNEURAL NETWORKS\n\nWenjie Wei1, Malu Zhang1∗, Zijian Zhou1, Ammar Belatreche2,\nYimeng Shan3, Yu Liang1, Honglin Cao1, Jieyuan Zhang1, Yang Yang1\n\n1University of Electronic Science and Technology of China\n2Northumbria University, 3Liaoning Technical University\n\nABSTRACT\... | true |
5 | <insight>
Achieving optimal outcomes when refining complex computational models involves navigating subtle trade-offs beyond simple scaling or reduction. While techniques like quantization and pruning aim to create more efficient, lightweight models by reducing complexity, they risk performance degradation if not caref... | -1,423.164063 | -1,514.804688 | -1,342.328125 | -1,586.558594 | 3,182.199219 | -10.165458 | -10.820033 | -9.588058 | -11.332561 | [
"QP-SNNS: QUANTIZED AND PRUNED SPIKING\nNEURAL NETWORKS\n\nWenjie Wei1, Malu Zhang1∗, Zijian Zhou1, Ammar Belatreche2,\nYimeng Shan3, Yu Liang1, Honglin Cao1, Jieyuan Zhang1, Yang Yang1\n\n1University of Electronic Science and Technology of China\n2Northumbria University, 3Liaoning Technical University\n\nABSTRACT\... | true |
5 | <insight>
Achieving both high performance and computational efficiency in neural networks requires navigating complexities beyond optimizing either aspect in isolation. Techniques designed to make models smaller and faster, such as reducing parameter precision or removing redundant components, can have their effectiven... | -1,406.414063 | -1,535.28125 | -1,420.25 | -1,464.265625 | 2,985.710938 | -10.045815 | -10.966294 | -10.144643 | -10.459041 | [
"QP-SNNS: QUANTIZED AND PRUNED SPIKING\nNEURAL NETWORKS\n\nWenjie Wei1, Malu Zhang1∗, Zijian Zhou1, Ammar Belatreche2,\nYimeng Shan3, Yu Liang1, Honglin Cao1, Jieyuan Zhang1, Yang Yang1\n\n1University of Electronic Science and Technology of China\n2Northumbria University, 3Liaoning Technical University\n\nABSTRACT\... | true |
6 | <insight>
Hierarchical approaches to complex sequential decision-making, which involve breaking down problems into layers (e.g., strategic resource allocation followed by tactical item selection), may implicitly interact with the underlying geometric structure of the problem's representation space, particularly when le... | -2,208.296875 | -1,622.65625 | -1,689.3125 | -2,165.492188 | 4,307.132813 | -12.618839 | -9.272322 | -9.653214 | -12.374241 | [
"SEQUENTIAL STOCHASTIC COMBINATORIAL OPTI-\nMIZATION USING HIERARCHICAL REINFORCEMENT\nLEARNING\n\nXinsong Feng1, Zihan Yu2, Yanhai Xiong3, Haipeng Chen3\n1UCLA, 2The University of Hong Kong, 3William & Mary\nxsfeng@ucla.edu, u3634664@connect.hku.hk, {yxiong05, hchen23}@wm.edu\n\nABSTRACT\n\nReinforcement learning ... | true |
6 | <insight>
Hierarchical approaches designed to tackle complex, sequential decision-making problems may gain significant advantages in stability and effectiveness by aligning their structure with the intrinsic geometry of the underlying state space. The observation that states necessitating similar optimal actions or yie... | -2,947.65625 | -2,272.28125 | -2,314.109375 | -2,977.59375 | 5,883.421875 | -14.309011 | -11.030492 | -11.233541 | -14.454339 | [
"SEQUENTIAL STOCHASTIC COMBINATORIAL OPTI-\nMIZATION USING HIERARCHICAL REINFORCEMENT\nLEARNING\n\nXinsong Feng1, Zihan Yu2, Yanhai Xiong3, Haipeng Chen3\n1UCLA, 2The University of Hong Kong, 3William & Mary\nxsfeng@ucla.edu, u3634664@connect.hku.hk, {yxiong05, hchen23}@wm.edu\n\nABSTRACT\n\nReinforcement learning ... | true |
6 | <insight>
Complex sequential decision-making problems, often tackled with hierarchical learning structures that separate high-level strategy (like resource allocation over time) from low-level execution (like specific actions at each step), may benefit significantly from inherent geometric properties within the problem... | -2,238.070313 | -2,423.820313 | -2,583.34375 | -2,592.84375 | 4,671.390625 | -12.097677 | -13.101731 | -13.964021 | -14.015371 | [
"SEQUENTIAL STOCHASTIC COMBINATORIAL OPTI-\nMIZATION USING HIERARCHICAL REINFORCEMENT\nLEARNING\n\nXinsong Feng1, Zihan Yu2, Yanhai Xiong3, Haipeng Chen3\n1UCLA, 2The University of Hong Kong, 3William & Mary\nxsfeng@ucla.edu, u3634664@connect.hku.hk, {yxiong05, hchen23}@wm.edu\n\nABSTRACT\n\nReinforcement learning ... | true |
7 | <insight>
Advanced intelligent systems face distinct but complementary challenges related to temporal dynamics. One challenge involves learning robustly from external feedback sources whose quality or characteristics evolve over time, such as adapting to a human supervisor who becomes more proficient. This necessitates... | -1,261.994141 | -1,375.148438 | -1,535.21875 | -1,447.386719 | 2,549.310547 | -9.417867 | -10.262301 | -11.456857 | -10.801394 | [
"LEARNING FROM IMPERFECT HUMAN FEEDBACK: A\nTALE FROM CORRUPTION-ROBUST DUELING\n\nYuwei Cheng\nDepartment of Statistics\nUniversity of Chicago\nChicago, IL 60637, USA\nyuweicheng@uchicago.edu\n\nXuefeng Liu∗\nDepartment of Computer Science\nUniversity of Chicago\nChicago, IL 60637, USA\nxuefeng@uchicago.edu\n\nFan... | true |
7 | <insight>
Designing machine learning systems often involves tackling distinct challenges: one is modeling complex, structured data, such as sequences exhibiting rich temporal dependencies, requiring powerful architectures. Another is learning effectively from imperfect or evolving training signals, like human feedback ... | -1,402.859375 | -1,512.921875 | -1,611.960938 | -1,649.603516 | 2,953.423828 | -10.315143 | -11.124426 | -11.852654 | -12.129437 | [
"LEARNING FROM IMPERFECT HUMAN FEEDBACK: A\nTALE FROM CORRUPTION-ROBUST DUELING\n\nYuwei Cheng\nDepartment of Statistics\nUniversity of Chicago\nChicago, IL 60637, USA\nyuweicheng@uchicago.edu\n\nXuefeng Liu∗\nDepartment of Computer Science\nUniversity of Chicago\nChicago, IL 60637, USA\nxuefeng@uchicago.edu\n\nFan... | true |
7 | <insight>
Developing advanced intelligent systems requires solving two fundamentally different types of problems regarding information fidelity. First, systems must learn effectively even when guidance signals, such as human preferences, are inherently flawed, inconsistent, or change dynamically over time (potentially ... | -1,831.515625 | -1,976.296875 | -2,004.96875 | -2,118.816406 | 3,921.660156 | -11.305652 | -12.199364 | -12.37635 | -13.079114 | [
"LEARNING FROM IMPERFECT HUMAN FEEDBACK: A\nTALE FROM CORRUPTION-ROBUST DUELING\n\nYuwei Cheng\nDepartment of Statistics\nUniversity of Chicago\nChicago, IL 60637, USA\nyuweicheng@uchicago.edu\n\nXuefeng Liu∗\nDepartment of Computer Science\nUniversity of Chicago\nChicago, IL 60637, USA\nxuefeng@uchicago.edu\n\nFan... | true |
8 | <insight>
The internal activation states within large language models represent a key leverage point for achieving distinct goals without costly retraining. It's possible to manipulate these activations during inference in multiple ways simultaneously. For instance, one can introduce sparsity by zeroing out less signif... | -1,320.015625 | -2,039.234375 | -1,498.109375 | -1,859.4375 | 3,720.578125 | -8.919024 | -13.77861 | -10.12236 | -12.563766 | [
"TRAINING-FREE ACTIVATION SPARSITY IN\nLARGE LANGUAGE MODELS\n\nJames Liu1,2∗ Pragaash Ponnusamy2 Tianle Cai3 Han Guo1 Yoon Kim1 Ben Athiwaratkun2\n\n1 Massachusetts Institute of Technology 2 Together AI\n\n3 Princeton University\n\n(cid:135) https://github.com/FasterDecoding/TEAL\n\nABSTRACT\n\nActivation sparsity... | true |
8 | <insight>
The internal activation states within large language models possess a structure that enables different kinds of beneficial, training-free manipulation during inference. Widespread computational efficiency can be achieved by zeroing out a significant fraction of low-magnitude activation values globally across ... | -1,440.5625 | -2,095.375 | -1,681.90625 | -1,938.617188 | 3,792.648438 | -9.234375 | -13.431891 | -10.78145 | -12.427033 | [
"TRAINING-FREE ACTIVATION SPARSITY IN\nLARGE LANGUAGE MODELS\n\nJames Liu1,2∗ Pragaash Ponnusamy2 Tianle Cai3 Han Guo1 Yoon Kim1 Ben Athiwaratkun2\n\n1 Massachusetts Institute of Technology 2 Together AI\n\n3 Princeton University\n\n(cid:135) https://github.com/FasterDecoding/TEAL\n\nABSTRACT\n\nActivation sparsity... | true |
8 | <insight>
The internal activation states within large language models possess a remarkable degree of malleability that can be exploited *during inference* without any need for retraining or fine-tuning. This flexibility allows for fundamentally different goals to be achieved through direct, targeted manipulation of the... | -1,436.1875 | -2,213.5 | -1,711.953125 | -2,075.859375 | 4,013.59375 | -8.757241 | -13.496951 | -10.438739 | -12.65768 | [
"TRAINING-FREE ACTIVATION SPARSITY IN\nLARGE LANGUAGE MODELS\n\nJames Liu1,2∗ Pragaash Ponnusamy2 Tianle Cai3 Han Guo1 Yoon Kim1 Ben Athiwaratkun2\n\n1 Massachusetts Institute of Technology 2 Together AI\n\n3 Princeton University\n\n(cid:135) https://github.com/FasterDecoding/TEAL\n\nABSTRACT\n\nActivation sparsity... | true |
10 | <insight>
Advancements in complex machine learning domains, whether dealing with structured data like directed graphs or large opaque models like language models accessed via APIs, increasingly rely on developing highly specialized techniques to inject critical contextual information. For graphs, particularly directed ... | -1,457.984375 | -2,062.140625 | -1,572.4375 | -1,880.039063 | 3,827.726563 | -8.890148 | -12.574028 | -9.588034 | -11.463653 | [
"WHAT ARE GOOD POSITIONAL ENCODINGS FOR\nDIRECTED GRAPHS?\n\nYinan Huang\nGeorgia Institute of Technology\nyhuang903@gatech.edu\n\nPan Li\nGeorgia Institute of Technology\npanli@gatech.edu\n\nHaoyu Wang\nGeorgia Institute of Technology\nhaoyu.wang@gatech.edu\n\nABSTRACT\n\nPositional encodings (PEs) are essential f... | true |
10 | <insight>
Effectively interfacing with complex computational systems, whether they are graph neural networks processing intricate directed structures or large language models accessed as black boxes, increasingly relies on the sophisticated design and optimization of specialized input representations. Techniques are be... | -1,818.234375 | -2,650.5625 | -2,168.453125 | -2,591.4375 | 4,891.78125 | -9.420903 | -13.733484 | -11.235508 | -13.427137 | [
"WHAT ARE GOOD POSITIONAL ENCODINGS FOR\nDIRECTED GRAPHS?\n\nYinan Huang\nGeorgia Institute of Technology\nyhuang903@gatech.edu\n\nPan Li\nGeorgia Institute of Technology\npanli@gatech.edu\n\nHaoyu Wang\nGeorgia Institute of Technology\nhaoyu.wang@gatech.edu\n\nABSTRACT\n\nPositional encodings (PEs) are essential f... | true |
10 | <insight>
Effectively leveraging complex computational models requires tailored approaches depending on whether the primary challenge lies in representing intricate inherent structure or in optimizing interaction under constraints. For systems with rich internal relationships, such as directed graphs, advancements focu... | -1,659.804688 | -2,881.375 | -2,164.417969 | -2,846.441406 | 5,223.203125 | -8.340727 | -14.479271 | -10.876472 | -14.303725 | [
"WHAT ARE GOOD POSITIONAL ENCODINGS FOR\nDIRECTED GRAPHS?\n\nYinan Huang\nGeorgia Institute of Technology\nyhuang903@gatech.edu\n\nPan Li\nGeorgia Institute of Technology\npanli@gatech.edu\n\nHaoyu Wang\nGeorgia Institute of Technology\nhaoyu.wang@gatech.edu\n\nABSTRACT\n\nPositional encodings (PEs) are essential f... | true |
11 | <insight>
Evaluating the sophisticated reasoning capabilities of advanced AI systems requires moving beyond traditional static benchmarks that primarily test final answer accuracy. A more comprehensive assessment emerges from combining two complementary approaches. First, rigorously testing the system's understanding *... | -2,658.890625 | -2,241.9375 | -2,756.828125 | -2,656.363281 | 4,800.363281 | -13.776635 | -11.616257 | -14.284083 | -13.76354 | [
"IS YOUR MODEL REALLY A GOOD MATH REASONER?\nEVALUATING MATHEMATICAL REASONING WITH\nCHECKLIST\n\nZihao Zhou12∗ Shudong Liu3∗ Maizhen Ning126 Wei Liu4\nDerek F. Wong3 Xiaowei Huang2 Qiufeng Wang1† Kaizhu Huang6\n1Xi’an Jiaotong-liverpool University\n4HKUST\nhttps://mathcheck.github.io/\n\n5William & Mary 6Duke Kuns... | true |
11 | <insight>Evaluating the reasoning abilities of advanced AI systems requires moving beyond traditional static benchmarks that primarily measure problem-solving accuracy. A more comprehensive assessment strategy is emerging, characterized by two complementary thrusts. One focuses on deepening the evaluation within specif... | -3,147.25 | -2,583.945313 | -3,208.1875 | -3,190.59375 | 5,713.601563 | -14.503456 | -11.907582 | -14.784274 | -14.703197 | [
"IS YOUR MODEL REALLY A GOOD MATH REASONER?\nEVALUATING MATHEMATICAL REASONING WITH\nCHECKLIST\n\nZihao Zhou12∗ Shudong Liu3∗ Maizhen Ning126 Wei Liu4\nDerek F. Wong3 Xiaowei Huang2 Qiufeng Wang1† Kaizhu Huang6\n1Xi’an Jiaotong-liverpool University\n4HKUST\nhttps://mathcheck.github.io/\n\n5William & Mary 6Duke Kuns... | true |
11 | <insight>
Evaluating the reasoning abilities of advanced AI models requires moving beyond simply checking if they get the right answer on static problems. Two complementary strategies are emerging to achieve a deeper assessment. One involves systematically testing models across a variety of related tasks and conditions... | -3,011.75 | -2,393.3125 | -3,068.359375 | -3,023.742188 | 5,360.445313 | -14.549517 | -11.561896 | -14.822992 | -14.60745 | [
"IS YOUR MODEL REALLY A GOOD MATH REASONER?\nEVALUATING MATHEMATICAL REASONING WITH\nCHECKLIST\n\nZihao Zhou12∗ Shudong Liu3∗ Maizhen Ning126 Wei Liu4\nDerek F. Wong3 Xiaowei Huang2 Qiufeng Wang1† Kaizhu Huang6\n1Xi’an Jiaotong-liverpool University\n4HKUST\nhttps://mathcheck.github.io/\n\n5William & Mary 6Duke Kuns... | true |
14 | <insight>
The way language models apply safety restrictions can be brittle, as demonstrated by their differing responses to harmful requests depending on whether the request is phrased in the present or past tense. This suggests that the models' internal grasp of semantic meaning may not consistently generalize across ... | -1,277.828125 | -1,318.84375 | -1,336.189453 | -1,242.361328 | 2,502.84375 | -8.935861 | -9.222684 | -9.343982 | -8.687841 | [
"DOES REFUSAL TRAINING IN LLMS GENERALIZE TO\nTHE PAST TENSE?\n\nMaksym Andriushchenko\nEPFL\n\nNicolas Flammarion\nEPFL\n\nABSTRACT\n\nRefusal training is widely used to prevent LLMs from generating harmful, unde-\nsirable, or illegal outputs. We reveal a curious generalization gap in the current\nrefusal training... | true |
14 | <insight>
Large language models can exhibit surprising brittleness in their safety training, where simple linguistic modifications, such as changing a request's verb tense from present to past (e.g., "How to do X?" vs. "How did people do X?"), can circumvent safety guardrails designed to prevent harmful outputs. This s... | -3,453.03125 | -3,412.75 | -3,528.8125 | -3,520.65625 | 6,857.625 | -14.151768 | -13.98668 | -14.462346 | -14.428919 | [
"DOES REFUSAL TRAINING IN LLMS GENERALIZE TO\nTHE PAST TENSE?\n\nMaksym Andriushchenko\nEPFL\n\nNicolas Flammarion\nEPFL\n\nABSTRACT\n\nRefusal training is widely used to prevent LLMs from generating harmful, unde-\nsirable, or illegal outputs. We reveal a curious generalization gap in the current\nrefusal training... | true |
14 | <insight>
The observation that large language models' safety refusals can often be circumvented by merely changing the tense of a request (e.g., from present to past) reveals a subtle brittleness in current alignment techniques. This vulnerability becomes particularly salient when considered alongside efforts to develo... | -1,437.804688 | -1,564.546875 | -1,550.09375 | -1,481.179688 | 2,933.4375 | -9.585364 | -10.430312 | -10.333959 | -9.874531 | [
"DOES REFUSAL TRAINING IN LLMS GENERALIZE TO\nTHE PAST TENSE?\n\nMaksym Andriushchenko\nEPFL\n\nNicolas Flammarion\nEPFL\n\nABSTRACT\n\nRefusal training is widely used to prevent LLMs from generating harmful, unde-\nsirable, or illegal outputs. We reveal a curious generalization gap in the current\nrefusal training... | true |
15 | <insight>
Neural architectures demonstrate a strong capacity to approximate complex, continuous mathematical operators, such as those mapping inputs to the solutions of differential equations, especially when the target operator is well-defined and potentially mirrors classical iterative solution structures. However, w... | -1,614.53125 | -1,532.1875 | -1,597.34375 | -1,801.007813 | 3,350.382813 | -9.610305 | -9.120164 | -9.507998 | -10.720284 | [
"JAILBREAKING AS A REWARD MISSPECIFICATION\nPROBLEM\n\nJiahui Gao1† Lei Li1 Zhenguo Li2 Qi Liu1 Lingpeng Kong1†\n\nZhihui Xie1\n1The University of Hong Kong\n{zhxieml,ggaojiahui,nlp.lilei}@gmail.com\n{li.zhenguo}@huawei.com\nWARNING: This paper contains examples of harmful language.\n\n{liuqi,lpk}@cs.hku.hk\n\n2Hua... | true |
15 | <insight>
Neural systems are trained via optimization to approximate complex transformations, mapping inputs to desired outputs. This optimization process can be remarkably effective, sometimes resulting in models whose internal structure implicitly learns or mirrors classical iterative algorithms used for solving rela... | -1,887.71875 | -1,426.117188 | -1,726.53125 | -2,067.539063 | 3,654.84375 | -10.545915 | -7.967135 | -9.645426 | -11.550498 | [
"JAILBREAKING AS A REWARD MISSPECIFICATION\nPROBLEM\n\nJiahui Gao1† Lei Li1 Zhenguo Li2 Qi Liu1 Lingpeng Kong1†\n\nZhihui Xie1\n1The University of Hong Kong\n{zhxieml,ggaojiahui,nlp.lilei}@gmail.com\n{li.zhenguo}@huawei.com\nWARNING: This paper contains examples of harmful language.\n\n{liuqi,lpk}@cs.hku.hk\n\n2Hua... | true |
15 | <insight>
Neural networks exhibit a remarkable ability to learn approximations of complex functions and operators, whether these represent solutions to well-defined mathematical problems like differential equations or more nuanced behavioral objectives like safety alignment. However, the success and reliability of thes... | -2,550.8125 | -2,559.09375 | -2,364.570313 | -3,410.71875 | 6,156.054688 | -11.286781 | -11.323423 | -10.462701 | -15.091676 | [
"JAILBREAKING AS A REWARD MISSPECIFICATION\nPROBLEM\n\nJiahui Gao1† Lei Li1 Zhenguo Li2 Qi Liu1 Lingpeng Kong1†\n\nZhihui Xie1\n1The University of Hong Kong\n{zhxieml,ggaojiahui,nlp.lilei}@gmail.com\n{li.zhenguo}@huawei.com\nWARNING: This paper contains examples of harmful language.\n\n{liuqi,lpk}@cs.hku.hk\n\n2Hua... | true |
17 | <insight>
The development of advanced AI systems reveals a notable contrast: while one strand of research meticulously analyzes the theoretical properties and fundamental limitations of powerful model classes, uncovering potential mathematical discontinuities or convergence issues when examined under specific theoretic... | -1,549.9375 | -1,902.0625 | -1,942.539063 | -1,568.4375 | 3,077.898438 | -10.13031 | -12.431781 | -12.696334 | -10.251225 | [
"HIGHER-ORDER GRAPHON NEURAL NETWORKS:\nAPPROXIMATION AND CUT DISTANCE\n\nDaniel Herbst∗ and Stefanie Jegelka∗,†,‡\n∗TUM, School of CIT †TUM, MCML and MDSI\n{daniel.herbst, stefanie.jegelka}@tum.de\n\n‡MIT, Department of EECS and CSAIL\n\nABSTRACT\n\nGraph limit models, like graphons for limits of dense graphs, hav... | true |
17 | <insight>
Advancing artificial intelligence capabilities involves tackling distinct types of challenges. On one hand, increasing the theoretical expressive power of models, particularly those designed for complex structured data, can lead to nuanced mathematical issues, such as discontinuities with respect to certain c... | -1,976.4375 | -2,574.875 | -2,617.53125 | -2,212.871094 | 4,146.652344 | -11.041551 | -14.384776 | -14.623079 | -12.362409 | [
"HIGHER-ORDER GRAPHON NEURAL NETWORKS:\nAPPROXIMATION AND CUT DISTANCE\n\nDaniel Herbst∗ and Stefanie Jegelka∗,†,‡\n∗TUM, School of CIT †TUM, MCML and MDSI\n{daniel.herbst, stefanie.jegelka}@tum.de\n\n‡MIT, Department of EECS and CSAIL\n\nABSTRACT\n\nGraph limit models, like graphons for limits of dense graphs, hav... | true |
17 | <insight>
Achieving cutting-edge performance in complex AI tasks, whether analyzing intricate relational structures or enabling seamless multimodal interaction, often involves a trade-off between theoretical expressivity and practical system stability and efficiency. While research continually pushes the boundaries of ... | -1,594.6875 | -2,143.671875 | -2,177.96875 | -1,803.835938 | 3,364.226563 | -9.904891 | -13.314733 | -13.527756 | -11.20395 | [
"HIGHER-ORDER GRAPHON NEURAL NETWORKS:\nAPPROXIMATION AND CUT DISTANCE\n\nDaniel Herbst∗ and Stefanie Jegelka∗,†,‡\n∗TUM, School of CIT †TUM, MCML and MDSI\n{daniel.herbst, stefanie.jegelka}@tum.de\n\n‡MIT, Department of EECS and CSAIL\n\nABSTRACT\n\nGraph limit models, like graphons for limits of dense graphs, hav... | true |
18 | <insight>
Significant advancements in AI performance, spanning both complex reasoning and robust real-world identification tasks (including handling previously unseen examples), can be unlocked by integrating fundamentally distinct types of information. Rather than relying solely on text or isolated data streams, power... | -1,984.640625 | -1,365.546875 | -2,031.703125 | -1,529.765625 | 2,848.25 | -13.319736 | -9.164744 | -13.635592 | -10.266884 | [
"THINK-ON-GRAPH 2.0: DEEP AND FAITHFUL LARGE\nLANGUAGE MODEL REASONING WITH KNOWLEDGE-\nGUIDED RETRIEVAL AUGMENTED GENERATION\n\nShengjie Ma1,2∗, Chengjin Xu1∗†, Xuhui Jiang1∗, Muzhi Li3, Huaren Qu4,\nCehao Yang1, Jiaxin Mao2†, Jian Guo1†\n1 IDEA Research, International Digital Economy Academy, Shenzhen, Guangdong,... | true |
18 | <insight>Advanced AI systems demonstrate significant performance gains by integrating diverse information sources tailored to the task at hand. For complex reasoning challenges, combining structured knowledge graphs (representing entities and relationships) with unstructured text documents through iterative, tightly-co... | -2,493.625 | -1,499 | -2,647.125 | -2,131.367188 | 3,476.867188 | -14.49782 | -8.715117 | -15.390262 | -12.391669 | [
"THINK-ON-GRAPH 2.0: DEEP AND FAITHFUL LARGE\nLANGUAGE MODEL REASONING WITH KNOWLEDGE-\nGUIDED RETRIEVAL AUGMENTED GENERATION\n\nShengjie Ma1,2∗, Chengjin Xu1∗†, Xuhui Jiang1∗, Muzhi Li3, Huaren Qu4,\nCehao Yang1, Jiaxin Mao2†, Jian Guo1†\n1 IDEA Research, International Digital Economy Academy, Shenzhen, Guangdong,... | true |
18 | <insight>
Advanced AI systems can achieve significantly enhanced capabilities, such as deeper reasoning and robust zero-shot classification, by moving beyond single information sources or modalities. Integrating diverse data types – for instance, by combining unstructured text with structured knowledge graphs through i... | -1,448.75 | -1,023.34375 | -1,478.382813 | -876.703125 | 1,870.414063 | -11.973141 | -8.457386 | -12.21804 | -7.245481 | [
"THINK-ON-GRAPH 2.0: DEEP AND FAITHFUL LARGE\nLANGUAGE MODEL REASONING WITH KNOWLEDGE-\nGUIDED RETRIEVAL AUGMENTED GENERATION\n\nShengjie Ma1,2∗, Chengjin Xu1∗†, Xuhui Jiang1∗, Muzhi Li3, Huaren Qu4,\nCehao Yang1, Jiaxin Mao2†, Jian Guo1†\n1 IDEA Research, International Digital Economy Academy, Shenzhen, Guangdong,... | true |
19 | <insight>
Architectures designed to enhance learning across distributed sources characterized by distinct data distributions (domains) often achieve success by explicitly capturing and utilizing domain-specific information, for instance, through specialized representations or adapters. However, this deliberate modeling... | -1,904.921875 | -2,594.171875 | -2,056.640625 | -3,007 | 5,449.453125 | -9.114459 | -12.412306 | -9.840385 | -14.38756 | [
"ENHANCING FEDERATED DOMAIN ADAPTATION WITH\nMULTI-DOMAIN PROTOTYPE-BASED FEDERATED FINE-\nTUNING\n\nJingyuan Zhang1, Yiyang Duan1, Shuaicheng Niu1, Yang Cao2, Wei Yang Bryan Lim1∗\n1 College of Computing and Data Science, Nanyang Technological University\n2 Department of Computer Science, Institute of Science Toky... | true |
19 | <insight>
Data that significantly deviates from the primary distribution (out-of-distribution, or OOD data) presents a dual challenge in distributed and evolving machine learning systems. On one hand, effectively integrating knowledge from diverse, mutually OOD sources (such as distinct domains in a federated network) ... | -1,433.59375 | -1,581.15625 | -1,620.5 | -2,004.621094 | 3,398.871094 | -8.688447 | -9.582766 | -9.821212 | -12.149219 | [
"ENHANCING FEDERATED DOMAIN ADAPTATION WITH\nMULTI-DOMAIN PROTOTYPE-BASED FEDERATED FINE-\nTUNING\n\nJingyuan Zhang1, Yiyang Duan1, Shuaicheng Niu1, Yang Cao2, Wei Yang Bryan Lim1∗\n1 College of Computing and Data Science, Nanyang Technological University\n2 Department of Computer Science, Institute of Science Toky... | true |
19 | <insight>
Managing data heterogeneity, particularly data points or groups that significantly differ from the main distribution (such as out-of-distribution samples or data from distinct domains in a federated system), presents fundamentally different challenges depending on *when* this heterogeneity is addressed. While... | -1,425.539063 | -2,118.734375 | -1,555.078125 | -2,448.296875 | 4,437.492188 | -7.705616 | -11.452619 | -8.405828 | -13.234037 | [
"ENHANCING FEDERATED DOMAIN ADAPTATION WITH\nMULTI-DOMAIN PROTOTYPE-BASED FEDERATED FINE-\nTUNING\n\nJingyuan Zhang1, Yiyang Duan1, Shuaicheng Niu1, Yang Cao2, Wei Yang Bryan Lim1∗\n1 College of Computing and Data Science, Nanyang Technological University\n2 Department of Computer Science, Institute of Science Toky... | true |
21 | <insight>
Advanced AI systems, particularly large language models, are being deployed to automate sophisticated tasks previously requiring human expertise. However, the strategies for effectively utilizing these AI systems vary significantly depending on the nature of the task. For tasks involving subjective judgment a... | -2,965.53125 | -2,962.28125 | -2,988.5625 | -3,080.28125 | 6,019.53125 | -14.326238 | -14.310537 | -14.4375 | -14.880586 | [
"DREAMBENCH++: A HUMAN-ALIGNED BENCHMARK\nFOR PERSONALIZED IMAGE GENERATION\n\nYuang Peng1,4,† Yuxin Cui1,† Haomiao Tang1,† Zekun Qi1 Runpei Dong2,¶\nJing Bai3 Chunrui Han4,‡ Zheng Ge4 Xiangyu Zhang4 Shu-Tao Xia1,¶\n\n1Tsinghua University\n\n2UIUC 3UCAS 4StepFun\n\nProject Page:\n\nDreamBench++\n\nFigure 1: Overvie... | true |
21 | <insight>
Advanced artificial intelligence models are increasingly capable of automating complex tasks traditionally requiring human expertise, but maximizing their effectiveness demands different strategies depending on the task's nature. For automating subjective evaluations, such as judging the quality or adherence ... | -3,015.21875 | -2,957.359375 | -3,011.640625 | -3,112.910156 | 6,073.847656 | -14.708385 | -14.426144 | -14.690929 | -15.184928 | [
"DREAMBENCH++: A HUMAN-ALIGNED BENCHMARK\nFOR PERSONALIZED IMAGE GENERATION\n\nYuang Peng1,4,† Yuxin Cui1,† Haomiao Tang1,† Zekun Qi1 Runpei Dong2,¶\nJing Bai3 Chunrui Han4,‡ Zheng Ge4 Xiangyu Zhang4 Shu-Tao Xia1,¶\n\n1Tsinghua University\n\n2UIUC 3UCAS 4StepFun\n\nProject Page:\n\nDreamBench++\n\nFigure 1: Overvie... | true |
21 | <insight>
Effectively leveraging advanced AI, such as large language models, for complex tasks requires distinct strategies tailored to the nature of the problem. When the goal involves subjective assessment that aligns with human perception, like evaluating the quality or relevance of creative generated content, the f... | -2,783.78125 | -2,279.484375 | -2,866.65625 | -2,580.71875 | 4,777.328125 | -14.202966 | -11.630022 | -14.625797 | -13.166932 | [
"DREAMBENCH++: A HUMAN-ALIGNED BENCHMARK\nFOR PERSONALIZED IMAGE GENERATION\n\nYuang Peng1,4,† Yuxin Cui1,† Haomiao Tang1,† Zekun Qi1 Runpei Dong2,¶\nJing Bai3 Chunrui Han4,‡ Zheng Ge4 Xiangyu Zhang4 Shu-Tao Xia1,¶\n\n1Tsinghua University\n\n2UIUC 3UCAS 4StepFun\n\nProject Page:\n\nDreamBench++\n\nFigure 1: Overvie... | true |
22 | <insight>
The internal representations learned by complex machine learning models, particularly large language models, may inherently possess sophisticated, non-Euclidean geometric structures. This underlying geometry could help explain the complex and sometimes counter-intuitive ways training data influences model pre... | -2,142.210938 | -3,013.210938 | -2,372.84375 | -3,131.234375 | 5,913.8125 | -9.737323 | -13.696413 | -10.785653 | -14.232883 | [
"NEURAL NETWORKS ON SYMMETRIC SPACES\nOF NONCOMPACT TYPE\n\nXuan Son Nguyen, Shuo Yang, Aymeric Histace\nETIS, UMR 8051, CY Cergy Paris University, ENSEA, CNRS, France\n{xuan-son.nguyen,shuo.yang,aymeric.histace}@ensea.fr\n\nABSTRACT\n\nRecent works have demonstrated promising performances of neural networks on\nhy... | true |
22 | <insight>
Developing machine learning models grounded in specific non-Euclidean geometries (like hyperbolic spaces or SPD manifolds) involves designing fundamental operations, such as distances and layers, that inherently respect the structure of these spaces. Simultaneously, efforts to understand large-scale models tr... | -1,689.382813 | -2,873.882813 | -1,958.578125 | -2,919.28125 | 5,523.96875 | -7.821217 | -13.305013 | -9.067492 | -13.515191 | [
"NEURAL NETWORKS ON SYMMETRIC SPACES\nOF NONCOMPACT TYPE\n\nXuan Son Nguyen, Shuo Yang, Aymeric Histace\nETIS, UMR 8051, CY Cergy Paris University, ENSEA, CNRS, France\n{xuan-son.nguyen,shuo.yang,aymeric.histace}@ensea.fr\n\nABSTRACT\n\nRecent works have demonstrated promising performances of neural networks on\nhy... | true |
22 | <insight>
Understanding the causal link between specific training data points and a large model's outputs involves complex techniques, often revealing a difference between data that *factually supports* an output and data that *causally influenced* it during training. This challenge of tracing influence might be intrin... | -1,953.15625 | -3,299.4375 | -2,093.546875 | -3,396.8125 | 6,555.859375 | -8.418777 | -14.221713 | -9.023909 | -14.641433 | [
"NEURAL NETWORKS ON SYMMETRIC SPACES\nOF NONCOMPACT TYPE\n\nXuan Son Nguyen, Shuo Yang, Aymeric Histace\nETIS, UMR 8051, CY Cergy Paris University, ENSEA, CNRS, France\n{xuan-son.nguyen,shuo.yang,aymeric.histace}@ensea.fr\n\nABSTRACT\n\nRecent works have demonstrated promising performances of neural networks on\nhy... | true |
23 | <insight>
Tackling complexity in machine learning often involves adopting multi-level or hierarchical representations, but the specific nature of this strategy is tailored to the source of the complexity. For instance, when dealing with information integrated from multiple, potentially incomplete sources, employing var... | -1,767.53125 | -2,670 | -2,226.609375 | -2,454.132813 | 4,665.054688 | -8.972239 | -13.5533 | -11.302586 | -12.457527 | [
"SIMPLE YET EFFECTIVE INCOMPLETE MULTI-VIEW\nCLUSTERING: SIMILARITY-LEVEL IMPUTATION AND\nINTRA-VIEW HYBRID-GROUP PROTOTYPE CONSTRUC-\nTION\n\nShengju Yu 1, Zhibin Dong 1, Siwei Wang 2∗, Pei Zhang 1, Yi Zhang 1, Xinwang Liu 1∗,\nNaiyang Guan 3, Tiejun Li 1∗, Yiu-ming Cheung 4\n1National University of Defense Techno... | true |
23 | <insight>Utilizing multi-resolution or hierarchical data representations emerges as a potent strategy adaptable to distinct machine learning challenges involving complex information structures. This technique can effectively address complexities arising from integrating potentially incomplete information across multipl... | -1,204.5 | -1,331.359375 | -1,119.796875 | -1,160.890625 | 2,576.953125 | -9.954545 | -11.00297 | -9.254519 | -9.594137 | [
"SIMPLE YET EFFECTIVE INCOMPLETE MULTI-VIEW\nCLUSTERING: SIMILARITY-LEVEL IMPUTATION AND\nINTRA-VIEW HYBRID-GROUP PROTOTYPE CONSTRUC-\nTION\n\nShengju Yu 1, Zhibin Dong 1, Siwei Wang 2∗, Pei Zhang 1, Yi Zhang 1, Xinwang Liu 1∗,\nNaiyang Guan 3, Tiejun Li 1∗, Yiu-ming Cheung 4\n1National University of Defense Techno... | true |
23 | <insight>
Developing sophisticated data representations that operate at multiple levels of granularity or scale is a powerful strategy for overcoming distinct challenges in information processing. Whether dealing with the problem of incomplete information fragmented across different sources or the difficulty of managin... | -1,762.0625 | -2,328.796875 | -1,834.421875 | -2,046.351563 | 4,302.789063 | -10.304459 | -13.618695 | -10.727613 | -11.966969 | [
"SIMPLE YET EFFECTIVE INCOMPLETE MULTI-VIEW\nCLUSTERING: SIMILARITY-LEVEL IMPUTATION AND\nINTRA-VIEW HYBRID-GROUP PROTOTYPE CONSTRUC-\nTION\n\nShengju Yu 1, Zhibin Dong 1, Siwei Wang 2∗, Pei Zhang 1, Yi Zhang 1, Xinwang Liu 1∗,\nNaiyang Guan 3, Tiejun Li 1∗, Yiu-ming Cheung 4\n1National University of Defense Techno... | true |
24 | <insight>
Developing agents capable of specific tasks after broad pre-training on unlabeled or reward-free experience highlights a fundamental divergence in approach. One strategy involves explicitly incorporating limited human guidance, such as multimodal instructions, using semi-supervised learning techniques to dire... | -1,346.6875 | -2,106.851563 | -1,537.6875 | -1,620.671875 | 3,536.523438 | -8.632612 | -13.505459 | -9.856971 | -10.388923 | [
"GROOT-2: WEAKLY SUPERVISED MULTIMODAL IN-\nSTRUCTION FOLLOWING AGENTS\n\nShaofei Cai1,2∗, Bowei Zhang3∗, Zihao Wang1,2, Haowei Lin1,2\nXiaojian Ma5, Anji Liu4, Yitao Liang1†\n1Institute for Artificial Intelligence, Peking University\n2School of Intelligence Science and Technology, Peking University\n3School of Ele... | true |
24 | <insight>
Achieving agent adaptability through large-scale interaction data requires distinct strategies depending on the desired form of generalization. One approach leverages a combination of unlabeled interaction data and a smaller set of explicit, human-provided guidance (like multimodal instructions) within a semi... | -2,095.09375 | -2,159.703125 | -2,161.9375 | -1,995.394531 | 4,088.253906 | -12.180778 | -12.556414 | -12.569404 | -11.601131 | [
"GROOT-2: WEAKLY SUPERVISED MULTIMODAL IN-\nSTRUCTION FOLLOWING AGENTS\n\nShaofei Cai1,2∗, Bowei Zhang3∗, Zihao Wang1,2, Haowei Lin1,2\nXiaojian Ma5, Anji Liu4, Yitao Liang1†\n1Institute for Artificial Intelligence, Peking University\n2School of Intelligence Science and Technology, Peking University\n3School of Ele... | true |
24 | <insight>
Leveraging large datasets of unlabeled agent experience to build adaptable agents presents a core challenge. Two distinct philosophies emerge for bridging the gap between general behavior learning and specific task execution. One approach involves augmenting the vast unlabeled data with a small, targeted set ... | -2,923.6875 | -2,890.460938 | -2,919.03125 | -2,845.492188 | 5,740.609375 | -14.331801 | -14.168926 | -14.308977 | -13.948491 | [
"GROOT-2: WEAKLY SUPERVISED MULTIMODAL IN-\nSTRUCTION FOLLOWING AGENTS\n\nShaofei Cai1,2∗, Bowei Zhang3∗, Zihao Wang1,2, Haowei Lin1,2\nXiaojian Ma5, Anji Liu4, Yitao Liang1†\n1Institute for Artificial Intelligence, Peking University\n2School of Intelligence Science and Technology, Peking University\n3School of Ele... | true |
25 | <insight>
Achieving robust and comprehensive outcomes in complex systems, whether representing intricate physical structures or aligning intelligent agents, benefits significantly from a strategy that moves beyond analyzing individual components in isolation. By explicitly modeling and leveraging the interactions and r... | -1,250.125 | -1,271.15625 | -1,298.640625 | -1,212.8125 | 2,435.453125 | -9.921627 | -10.088542 | -10.306672 | -9.625496 | [
"POLYHEDRONNET: REPRESENTATION LEARNING FOR\nPOLYHEDRA WITH SURFACE-ATTRIBUTED GRAPH\n\nDazhou Yu Genpei Zhang Liang Zhao ∗\nDepartment of Computer Science, Emory University\ndyu62@emory.edu, genpeizhang2024@gmail.com, liang.zhao@emory.edu\n\nABSTRACT\n\nUbiquitous geometric objects can be precisely and efficiently... | true |
25 | <insight>
Advanced learning systems face challenges whether modeling complex physical structures or aligning abstract behaviors. Two distinct but potentially complementary strategies emerge for handling such complexity. One involves explicitly modeling the intricate relationships between components (like the vertices, ... | -2,411.882813 | -2,398.53125 | -1,992.625 | -2,827.011719 | 5,644.800781 | -11.765282 | -11.700152 | -9.720122 | -13.790301 | [
"POLYHEDRONNET: REPRESENTATION LEARNING FOR\nPOLYHEDRA WITH SURFACE-ATTRIBUTED GRAPH\n\nDazhou Yu Genpei Zhang Liang Zhao ∗\nDepartment of Computer Science, Emory University\ndyu62@emory.edu, genpeizhang2024@gmail.com, liang.zhao@emory.edu\n\nABSTRACT\n\nUbiquitous geometric objects can be precisely and efficiently... | true |
25 | <insight>
Effectively modeling complex systems, whether they are intricate geometric objects or sophisticated artificial intelligences, often necessitates moving beyond analyzing isolated components. A powerful approach involves decomposing the system into meaningful sub-units – such as local geometric regions defined ... | -1,478.578125 | -1,436.890625 | -1,404.671875 | -1,613.960938 | 3,124.757813 | -9.791908 | -9.515832 | -9.302463 | -10.688483 | [
"POLYHEDRONNET: REPRESENTATION LEARNING FOR\nPOLYHEDRA WITH SURFACE-ATTRIBUTED GRAPH\n\nDazhou Yu Genpei Zhang Liang Zhao ∗\nDepartment of Computer Science, Emory University\ndyu62@emory.edu, genpeizhang2024@gmail.com, liang.zhao@emory.edu\n\nABSTRACT\n\nUbiquitous geometric objects can be precisely and efficiently... | true |
26 | <insight>
Current methods for training large AI models often involve sophisticated optimization techniques designed to navigate complex "loss landscapes," sometimes visualized as deep valleys with rivers at the bottom. These techniques use specific learning rate schedules with distinct phases (e.g., a stable phase for ... | -3,147.78125 | -3,183.3125 | -3,238.28125 | -3,251.96875 | 6,344.78125 | -14.115611 | -14.274944 | -14.521441 | -14.58282 | [
"THE LABYRINTH OF LINKS: NAVIGATING THE ASSO-\nCIATIVE MAZE OF MULTI-MODAL LLMS\n\nHong Li, Nanxi Li, Yuanjie Chen, Jianbin Zhu, Qinlu Guo, Cewu Lu, Yong-Lu Li∗\nShanghai Jiao Tong University, Shanghai Innovation Institute\n{hong li,andyc 03,cyj2003,bin pig,guoqinlu,lucewu,yonglu li}@sjtu.edu.cn\n\nABSTRACT\n\nMult... | true |
26 | <insight>
Advanced training techniques are being developed to optimize large models efficiently, navigating complex learning landscapes in ways where standard progress metrics like training loss can be misleading during certain phases. These methods allow for flexible training runs and effective optimization even under... | -2,158.019531 | -2,134.148438 | -2,172.617188 | -2,102.242188 | 4,221.792969 | -12.402411 | -12.265221 | -12.486305 | -12.081852 | [
"THE LABYRINTH OF LINKS: NAVIGATING THE ASSO-\nCIATIVE MAZE OF MULTI-MODAL LLMS\n\nHong Li, Nanxi Li, Yuanjie Chen, Jianbin Zhu, Qinlu Guo, Cewu Lu, Yong-Lu Li∗\nShanghai Jiao Tong University, Shanghai Innovation Institute\n{hong li,andyc 03,cyj2003,bin pig,guoqinlu,lucewu,yonglu li}@sjtu.edu.cn\n\nABSTRACT\n\nMult... | true |
26 | <insight>
The training process for large models often involves distinct phases, such as an extended period with a relatively high learning rate followed by a rapid decay. This approach, while effective for navigating the general landscape of solutions and making broad progress (akin to moving along a 'river valley'), m... | -2,460.882813 | -2,452.421875 | -2,493.8125 | -2,389.859375 | 4,809.351563 | -13.230553 | -13.185064 | -13.407594 | -12.848706 | [
"THE LABYRINTH OF LINKS: NAVIGATING THE ASSO-\nCIATIVE MAZE OF MULTI-MODAL LLMS\n\nHong Li, Nanxi Li, Yuanjie Chen, Jianbin Zhu, Qinlu Guo, Cewu Lu, Yong-Lu Li∗\nShanghai Jiao Tong University, Shanghai Innovation Institute\n{hong li,andyc 03,cyj2003,bin pig,guoqinlu,lucewu,yonglu li}@sjtu.edu.cn\n\nABSTRACT\n\nMult... | true |
27 | <insight>
Effective machine learning across diverse data types, such as sequential text and structured tables, benefits from moving beyond uniform treatment of input features or tokens. Standard evaluation metrics that average performance across all input parts can obscure weaknesses in handling critical information, w... | -2,964.3125 | -2,910.125 | -2,979.5625 | -3,002.757813 | 5,897.632813 | -13.723669 | -13.472801 | -13.794271 | -13.901656 | [
"WHAT IS WRONG WITH PERPLEXITY FOR LONG-\nCONTEXT LANGUAGE MODELING?\n\nLizhe Fang1∗ Yifei Wang2∗ Zhaoyang Liu3 Chenheng Zhang1\nStefanie Jegelka4,5\n1 State Key Lab of General Artificial Intelligence,\n\nJinyang Gao3 Bolin Ding3 Yisen Wang1,6†\n\nSchool of Intelligence Science and Technology, Peking University\n\n... | true |
27 | <insight>
Improving machine learning model capabilities, whether for processing long sequences or structured tabular data, can be significantly advanced by shifting focus from global properties or external manipulations (like data augmentation) towards leveraging the *internal structure and relationships within individ... | -2,651.71875 | -2,657.890625 | -2,730.433594 | -2,627.724609 | 5,206.900391 | -13.460502 | -13.491831 | -13.860069 | -13.338703 | [
"WHAT IS WRONG WITH PERPLEXITY FOR LONG-\nCONTEXT LANGUAGE MODELING?\n\nLizhe Fang1∗ Yifei Wang2∗ Zhaoyang Liu3 Chenheng Zhang1\nStefanie Jegelka4,5\n1 State Key Lab of General Artificial Intelligence,\n\nJinyang Gao3 Bolin Ding3 Yisen Wang1,6†\n\nSchool of Intelligence Science and Technology, Peking University\n\n... | true |
27 | <insight>
Across different data types, such as long sequences of text and structured tabular data, relying on evaluation or training methods that treat all input components uniformly (e.g., averaging performance metrics across all tokens or features) can be fundamentally misleading. Such approaches often mask a model's... | -2,595.53125 | -2,530.28125 | -2,615.050781 | -2,581.765625 | 5,092.527344 | -13.310416 | -12.975801 | -13.410517 | -13.239823 | [
"WHAT IS WRONG WITH PERPLEXITY FOR LONG-\nCONTEXT LANGUAGE MODELING?\n\nLizhe Fang1∗ Yifei Wang2∗ Zhaoyang Liu3 Chenheng Zhang1\nStefanie Jegelka4,5\n1 State Key Lab of General Artificial Intelligence,\n\nJinyang Gao3 Bolin Ding3 Yisen Wang1,6†\n\nSchool of Intelligence Science and Technology, Peking University\n\n... | true |
28 | <insight>
Improving the reliability and appropriateness of generative AI outputs can be achieved dynamically during inference time through distinct but potentially complementary mechanisms, avoiding the need for retraining. One approach involves a form of self-correction where a secondary generative process (e.g., gene... | -2,461.6875 | -2,625.625 | -2,496.828125 | -2,615.039063 | 5,205.523438 | -12.55963 | -13.396046 | -12.738919 | -13.342036 | [
"SELF-CORRECTING DECODING WITH GENERATIVE\nFEEDBACK FOR MITIGATING HALLUCINATIONS IN\nLARGE VISION-LANGUAGE MODELS\n\nCe Zhang∗1 Zifu Wan∗1 Zhehan Kan2 Martin Q. Ma1 Simon Stepputtis1\nDeva Ramanan1 Russ Salakhutdinov1 Louis-Philippe Morency1 Katia Sycara1 Yaqi Xie1\n1School of Computer Science, Carnegie Mellon Uni... | true |
28 | <insight>
Achieving reliable and adaptable AI outputs requires controlling different aspects of generation at inference time. One critical aspect is ensuring factual grounding, particularly in multimodal settings where text must align with visual information. This can be addressed using internal consistency checks, suc... | -2,918.460938 | -3,123.484375 | -3,006.96875 | -3,168.78125 | 6,203.757813 | -13.387436 | -14.32791 | -13.793435 | -14.535694 | [
"SELF-CORRECTING DECODING WITH GENERATIVE\nFEEDBACK FOR MITIGATING HALLUCINATIONS IN\nLARGE VISION-LANGUAGE MODELS\n\nCe Zhang∗1 Zifu Wan∗1 Zhehan Kan2 Martin Q. Ma1 Simon Stepputtis1\nDeva Ramanan1 Russ Salakhutdinov1 Louis-Philippe Morency1 Katia Sycara1 Yaqi Xie1\n1School of Computer Science, Carnegie Mellon Uni... | true |
28 | <insight>
There are distinct, yet complementary, approaches emerging to enhance the reliability and adaptability of complex AI models (like those processing language and vision) *at inference time*, bypassing the need for costly retraining. One approach focuses on self-correction by leveraging the model's ecosystem: it... | -2,682.15625 | -2,824.945313 | -2,750.015625 | -2,889.894531 | 5,646.980469 | -13.020176 | -13.713326 | -13.34959 | -14.028614 | [
"SELF-CORRECTING DECODING WITH GENERATIVE\nFEEDBACK FOR MITIGATING HALLUCINATIONS IN\nLARGE VISION-LANGUAGE MODELS\n\nCe Zhang∗1 Zifu Wan∗1 Zhehan Kan2 Martin Q. Ma1 Simon Stepputtis1\nDeva Ramanan1 Russ Salakhutdinov1 Louis-Philippe Morency1 Katia Sycara1 Yaqi Xie1\n1School of Computer Science, Carnegie Mellon Uni... | true |
29 | <insight>
Advanced learning techniques can uncover the underlying rules governing how systems change over time by learning sophisticated internal representations from data, particularly leveraging temporal structure. However, practical deployment often necessitates simplifying these models, for instance, by using lower... | -1,595.3125 | -1,322.859375 | -1,576.546875 | -1,677.578125 | 3,019.203125 | -10.635417 | -8.819062 | -10.510312 | -11.183854 | [
"SCALING LAWS FOR PRECISION\n\nTanishq Kumar∗ 1 Zachary Ankner* 3,4 Benjamin F. Spector2 Blake Bordelon1\nNiklas Muennighoff2 Mansheej Paul4 Cengiz Pehlevan1 Christopher R´e2\nAditi Raghunathan5\n\n1Harvard University\n4Databricks\n3MIT\n\n2Stanford University\n5Carnegie Mellon University\n\nABSTRACT\n\nLow precisi... | true |
29 | <insight>
Techniques designed to uncover complex, underlying system dynamics by learning from vast amounts of data, particularly leveraging temporal or structural patterns, might inadvertently create models whose performance is exceptionally sensitive to the numerical precision used. While reducing precision (e.g., usi... | -1,605.3125 | -1,368.640625 | -1,686.453125 | -1,865.691406 | 3,153.191406 | -10.033203 | -8.554004 | -10.540332 | -11.660571 | [
"SCALING LAWS FOR PRECISION\n\nTanishq Kumar∗ 1 Zachary Ankner* 3,4 Benjamin F. Spector2 Blake Bordelon1\nNiklas Muennighoff2 Mansheej Paul4 Cengiz Pehlevan1 Christopher R´e2\nAditi Raghunathan5\n\n1Harvard University\n4Databricks\n3MIT\n\n2Stanford University\n5Carnegie Mellon University\n\nABSTRACT\n\nLow precisi... | true |
29 | <insight>
Learning intricate temporal patterns or underlying system dynamics from large datasets can lead to representations that are highly effective but also potentially fragile. Reducing the numerical precision of these representations, a common technique for improving computational efficiency (especially during inf... | -1,097.65625 | -1,084.796875 | -1,298.375 | -1,247.070313 | 2,131.148438 | -8.924034 | -8.819487 | -10.555894 | -10.138783 | [
"SCALING LAWS FOR PRECISION\n\nTanishq Kumar∗ 1 Zachary Ankner* 3,4 Benjamin F. Spector2 Blake Bordelon1\nNiklas Muennighoff2 Mansheej Paul4 Cengiz Pehlevan1 Christopher R´e2\nAditi Raghunathan5\n\n1Harvard University\n4Databricks\n3MIT\n\n2Stanford University\n5Carnegie Mellon University\n\nABSTRACT\n\nLow precisi... | true |
30 | <insight>
Enhancing the ability of sequence models to handle extremely long inputs, for instance by optimizing internal memory mechanisms to selectively retain only critical information over vast distances, addresses a key bottleneck in processing extended contexts. However, successfully retaining information across lo... | -2,091.390625 | -1,918.835938 | -1,928.546875 | -2,155.765625 | 4,237.445313 | -12.752382 | -11.700219 | -11.759432 | -13.144913 | [
"LONGMAMBA: ENHANCING MAMBA’S LONG CON-\nTEXT CAPABILITIES VIA TRAINING-FREE RECEPTIVE\nFIELD ENLARGEMENT\n\nZhifan Ye1∗, Kejing Xia1∗, Yonggan Fu1,2, Xin Dong2, Jihoon Hong1, Xiangchi Yuan1,\nShizhe Diao2, Jan Kautz2, Pavlo Molchanov2, Yingyan (Celine) Lin1,2\n1Georgia Institute of Technology 2NVIDIA\n{zye327,kxia... | true |
30 | <insight>
Advanced AI models face distinct challenges when operating at the extremes of either sequence length or computational complexity. While architectures optimized for efficiently processing very long sequences can falter due to mechanisms like memory decay hindering the retention of critical information over vas... | -2,172.953125 | -2,013.164063 | -2,044.257813 | -2,195.441406 | 4,337.300781 | -13.090079 | -12.127495 | -12.314806 | -13.225551 | [
"LONGMAMBA: ENHANCING MAMBA’S LONG CON-\nTEXT CAPABILITIES VIA TRAINING-FREE RECEPTIVE\nFIELD ENLARGEMENT\n\nZhifan Ye1∗, Kejing Xia1∗, Yonggan Fu1,2, Xin Dong2, Jihoon Hong1, Xiangchi Yuan1,\nShizhe Diao2, Jan Kautz2, Pavlo Molchanov2, Yingyan (Celine) Lin1,2\n1Georgia Institute of Technology 2NVIDIA\n{zye327,kxia... | true |
31 | <insight>
The way internal data representations are structured and utilized within deep learning models can be more critical to performance than simply increasing model complexity or relying on raw output features. Specifically, deliberately structuring an intermediate representation, such as by breaking it down into d... | -1,565.46875 | -2,321.4375 | -1,595.4375 | -2,257.273438 | 4,548.742188 | -8.945536 | -13.265357 | -9.116786 | -12.898705 | [
"DON’T FLATTEN, TOKENIZE! UNLOCKING THE KEY TO\nSOFTMOE’S EFFICACY IN DEEP RL\n\nGhada Sokar\nGoogle DeepMind\ngsokar@google.com\n\nJohan Obando-Ceron\nMila, Universit´e de Montr´eal\njobando0730@gmail.com\n\nAaron Courville\nMila, Universit´e de Montr´eal\ncourvila@mila.quebec\n\nHugo Larochelle\nGoogle DeepMind\n... | true |
31 | <insight>
Deep learning models across different domains, such as sequential decision-making and data typicality assessment, may achieve significant performance and robustness gains by operating on more structured or abstracted internal representations rather than directly on dense, continuous ones. Specifically, transf... | -1,132.03125 | -1,331.1875 | -1,286.5 | -1,135.668945 | 2,312.387695 | -9.278945 | -10.911373 | -10.545082 | -9.308762 | [
"DON’T FLATTEN, TOKENIZE! UNLOCKING THE KEY TO\nSOFTMOE’S EFFICACY IN DEEP RL\n\nGhada Sokar\nGoogle DeepMind\ngsokar@google.com\n\nJohan Obando-Ceron\nMila, Universit´e de Montr´eal\njobando0730@gmail.com\n\nAaron Courville\nMila, Universit´e de Montr´eal\ncourvila@mila.quebec\n\nHugo Larochelle\nGoogle DeepMind\n... | true |
31 | <insight>
Structuring or analyzing the internal representations learned by deep networks appears to be a critical factor for enhancing performance across different challenging domains, potentially more impactful than simply scaling model components or focusing solely on raw input fidelity. Whether achieved through expl... | -1,121.859375 | -1,212.65625 | -1,102.703125 | -948.367188 | 2,180.179688 | -9.840872 | -10.637336 | -9.672834 | -8.319011 | [
"DON’T FLATTEN, TOKENIZE! UNLOCKING THE KEY TO\nSOFTMOE’S EFFICACY IN DEEP RL\n\nGhada Sokar\nGoogle DeepMind\ngsokar@google.com\n\nJohan Obando-Ceron\nMila, Universit´e de Montr´eal\njobando0730@gmail.com\n\nAaron Courville\nMila, Universit´e de Montr´eal\ncourvila@mila.quebec\n\nHugo Larochelle\nGoogle DeepMind\n... | true |
32 | <insight>
While advanced techniques are emerging to effectively guide large pre-trained generative models to specialize their outputs for specific target domains or styles with improved quality, a significant capability gap remains when applying these models to complex, real-world procedural tasks. Successfully adaptin... | -1,774.28125 | -1,681.625 | -1,730.640625 | -1,592.96875 | 3,318.234375 | -12.494939 | -11.842429 | -12.18761 | -11.21809 | [
"SPIDER 2.0: EVALUATING LANGUAGE MODELS ON\nREAL-WORLD ENTERPRISE TEXT-TO-SQL WORK-\nFLOWS\n\nFangyu Lei∗ ♠ Jixuan Chen∗♠ Yuxiao Ye♠ Ruisheng Cao♠ Dongchan Shin♠\nHongjin Su♠ Zhaoqing Suo♠ Hongcheng Gao♠ Wenjing Hu♠ Pengcheng Yin♡\nVictor Zhong⋆ Caiming Xiong♢ Ruoxi Sun△ Qian Liu♣ Sida I. Wang Tao Yu♠\n♠University ... | true |
32 | <insight>
While advanced AI models demonstrate impressive capabilities on standardized benchmarks, their performance often falters significantly when applied to complex, real-world tasks involving specialized domains, intricate constraints, and extensive context (such as enterprise software development or data analysis... | -2,767.09375 | -2,743.148438 | -2,786.34375 | -2,888.34375 | 5,612.242188 | -14.337274 | -14.213204 | -14.437015 | -14.965511 | [
"SPIDER 2.0: EVALUATING LANGUAGE MODELS ON\nREAL-WORLD ENTERPRISE TEXT-TO-SQL WORK-\nFLOWS\n\nFangyu Lei∗ ♠ Jixuan Chen∗♠ Yuxiao Ye♠ Ruisheng Cao♠ Dongchan Shin♠\nHongjin Su♠ Zhaoqing Suo♠ Hongcheng Gao♠ Wenjing Hu♠ Pengcheng Yin♡\nVictor Zhong⋆ Caiming Xiong♢ Ruoxi Sun△ Qian Liu♣ Sida I. Wang Tao Yu♠\n♠University ... | true |
32 | <insight>
While advanced AI models demonstrate impressive capabilities on standardized benchmarks, their performance often degrades significantly when confronted with the complexities of real-world enterprise workflows, such as intricate database interactions involving vast schemas, specific system dialects, and multi-... | -2,271.75 | -1,441.65625 | -1,301.3125 | -1,640.40625 | 4,052.5 | -14.378164 | -9.124407 | -8.236156 | -10.382318 | [
"SPIDER 2.0: EVALUATING LANGUAGE MODELS ON\nREAL-WORLD ENTERPRISE TEXT-TO-SQL WORK-\nFLOWS\n\nFangyu Lei∗ ♠ Jixuan Chen∗♠ Yuxiao Ye♠ Ruisheng Cao♠ Dongchan Shin♠\nHongjin Su♠ Zhaoqing Suo♠ Hongcheng Gao♠ Wenjing Hu♠ Pengcheng Yin♡\nVictor Zhong⋆ Caiming Xiong♢ Ruoxi Sun△ Qian Liu♣ Sida I. Wang Tao Yu♠\n♠University ... | true |
33 | <insight>
Building more capable and robust AI systems requires addressing distinct challenges at different processing stages, often necessitating fundamentally different architectural solutions. On one hand, ensuring accurate and reliable perception, especially when faced with novel or unexpected sensory details (like ... | -2,449.148438 | -2,620.757813 | -2,657.46875 | -2,578.121094 | 4,990.558594 | -12.432225 | -13.303339 | -13.489689 | -13.086909 | [
"BRAIN BANDIT: A BIOLOGICALLY GROUNDED NEU-\nRAL NETWORK FOR EFFICIENT CONTROL OF EXPLO-\nRATION\n\nChen Jiang1∗, Jiahui An2,1, Yating Liu3,2,1, Ni ji2,1†\n1Chinese Institute for Brain Research, Beijing\n2Chinese Academy of Medical Sciences & Peking Union Medical College\n3China Agricultural University\nchen.jiang3... | true |
33 | <insight>
Effective intelligent systems must navigate uncertainty at multiple levels. One critical challenge lies in accurately perceiving the environment, especially distinguishing fine-grained details like orientation or quantity, which standard feature encoders might overlook. Generative feedback mechanisms can enha... | -2,101.734375 | -2,375.585938 | -2,431.46875 | -2,324.65625 | 4,370.507813 | -11.179439 | -12.636095 | -12.933345 | -12.365192 | [
"BRAIN BANDIT: A BIOLOGICALLY GROUNDED NEU-\nRAL NETWORK FOR EFFICIENT CONTROL OF EXPLO-\nRATION\n\nChen Jiang1∗, Jiahui An2,1, Yating Liu3,2,1, Ni ji2,1†\n1Chinese Institute for Brain Research, Beijing\n2Chinese Academy of Medical Sciences & Peking Union Medical College\n3China Agricultural University\nchen.jiang3... | true |
33 | <insight>
Developing more capable AI systems involves tackling distinct challenges in both perception and action. One key challenge lies in enabling agents to make efficient decisions under uncertainty, particularly balancing exploration of the unknown with exploitation of known options. Progress here involves designin... | -3,216.632813 | -3,399.8125 | -3,458.59375 | -3,453.96875 | 6,611.820313 | -13.347024 | -14.107106 | -14.351011 | -14.33182 | [
"BRAIN BANDIT: A BIOLOGICALLY GROUNDED NEU-\nRAL NETWORK FOR EFFICIENT CONTROL OF EXPLO-\nRATION\n\nChen Jiang1∗, Jiahui An2,1, Yating Liu3,2,1, Ni ji2,1†\n1Chinese Institute for Brain Research, Beijing\n2Chinese Academy of Medical Sciences & Peking Union Medical College\n3China Agricultural University\nchen.jiang3... | true |
34 | <insight>
Approaches for handling complex structural data, whether abstract relational structures like hypergraphs or geometric structures like 3D curves, reveal a fundamental design choice regarding how structural information influences downstream tasks or generative processes. One strategy involves pre-processing and... | -2,859.1875 | -1,892 | -1,998.96875 | -2,872.789063 | 5,625.007813 | -13.236979 | -8.759259 | -9.254485 | -13.29995 | [
"TRAINING-FREE MESSAGE PASSING\nFOR LEARNING ON HYPERGRAPHS\n\nBohan Tang1 Zexi Liu2∗ Keyue Jiang3∗ Siheng Chen2,4 Xiaowen Dong1\n1University of Oxford\n2Shanghai Jiao Tong University\n4Shanghai AI Laboratory\nbohan.tang@eng.ox.ac.uk\n\n3University College London\n\nABSTRACT\n\nHypergraphs are crucial for modelling... | true |
34 | <insight>
There are distinct strategies emerging for integrating complex structural information (like higher-order network relationships or 3D geometric forms) with machine learning models, particularly concerning computational efficiency and optimization pathways. One approach significantly accelerates the training ph... | -2,416.234375 | -1,617.757813 | -1,621.046875 | -2,377.474609 | 4,790.419922 | -12.327726 | -8.253866 | -8.270647 | -12.129972 | [
"TRAINING-FREE MESSAGE PASSING\nFOR LEARNING ON HYPERGRAPHS\n\nBohan Tang1 Zexi Liu2∗ Keyue Jiang3∗ Siheng Chen2,4 Xiaowen Dong1\n1University of Oxford\n2Shanghai Jiao Tong University\n4Shanghai AI Laboratory\nbohan.tang@eng.ox.ac.uk\n\n3University College London\n\nABSTRACT\n\nHypergraphs are crucial for modelling... | true |
34 | <insight>
Significant advancements in handling complex data structures, like higher-order relationships or 3D geometry, can be achieved by strategically shifting or translating the computational burden, rather than solely optimizing the core learning algorithm itself. One effective approach is to decouple structural pr... | -2,259.230469 | -1,593.359375 | -1,402.234375 | -2,227.777344 | 4,678.132813 | -12.081447 | -8.520638 | -7.49858 | -11.913248 | [
"TRAINING-FREE MESSAGE PASSING\nFOR LEARNING ON HYPERGRAPHS\n\nBohan Tang1 Zexi Liu2∗ Keyue Jiang3∗ Siheng Chen2,4 Xiaowen Dong1\n1University of Oxford\n2Shanghai Jiao Tong University\n4Shanghai AI Laboratory\nbohan.tang@eng.ox.ac.uk\n\n3University College London\n\nABSTRACT\n\nHypergraphs are crucial for modelling... | true |
35 | <insight>
Large language models develop internal representations where complex, high-level concepts, including subjective viewpoints like political perspectives, are encoded in a surprisingly simple, linear fashion within specific components like attention head activations. This underlying linear structure within the m... | -1,426.203125 | -1,210.265625 | -1,301.125 | -1,174.21875 | 2,509.5625 | -10.187165 | -8.644754 | -9.29375 | -8.387277 | [
"SCALING DIFFUSION LANGUAGE MODELS\nVIA ADAPTATION FROM AUTOREGRESSIVE MODELS\n\n∗\n, Shivam Agarwal∗2, Yizhe Zhang3, Jiacheng Ye1, Lin Zheng1\n\nShansan Gong∗1\nMukai Li1, Chenxin An1, Peilin Zhao4, Wei Bi4, Hao Peng2, Jiawei Han, Lingpeng Kong1\n1The University of Hong Kong 2 University of Illinois at Urbana-Cham... | true |
35 | <insight>
The ability to efficiently transform large language models from one generative paradigm (e.g., sequential prediction) to another (e.g., iterative refinement) via adaptation suggests a deep transferability of learned knowledge. This transfer goes beyond surface linguistic fluency. Investigations into the inter... | -2,162.34375 | -1,783.320313 | -2,175.5 | -1,777.273438 | 3,547.4375 | -13.77289 | -11.358728 | -13.856688 | -11.320213 | [
"SCALING DIFFUSION LANGUAGE MODELS\nVIA ADAPTATION FROM AUTOREGRESSIVE MODELS\n\n∗\n, Shivam Agarwal∗2, Yizhe Zhang3, Jiacheng Ye1, Lin Zheng1\n\nShansan Gong∗1\nMukai Li1, Chenxin An1, Peilin Zhao4, Wei Bi4, Hao Peng2, Jiawei Han, Lingpeng Kong1\n1The University of Hong Kong 2 University of Illinois at Urbana-Cham... | true |
35 | <insight>
The extensive pre-training of large language models creates internal representations that are remarkably versatile. Not only can the models' overall capabilities be successfully repurposed for fundamentally different generative processes, such as converting sequence-by-sequence generation to a diffusion-based... | -1,070.21875 | -960.609375 | -1,133.8125 | -782.226563 | 1,679.242188 | -10.596225 | -9.510983 | -11.225866 | -7.744817 | [
"SCALING DIFFUSION LANGUAGE MODELS\nVIA ADAPTATION FROM AUTOREGRESSIVE MODELS\n\n∗\n, Shivam Agarwal∗2, Yizhe Zhang3, Jiacheng Ye1, Lin Zheng1\n\nShansan Gong∗1\nMukai Li1, Chenxin An1, Peilin Zhao4, Wei Bi4, Hao Peng2, Jiawei Han, Lingpeng Kong1\n1The University of Hong Kong 2 University of Illinois at Urbana-Cham... | true |
36 | <insight>
Achieving robust and efficient learning in complex, temporally extended scenarios requires addressing limitations at two distinct levels: the overall network structure's adaptability and the individual units' capacity for processing long-range temporal dependencies. While dynamically adjusting network topolog... | -3,157.359375 | -3,220.3125 | -3,197.796875 | -3,323.34375 | 6,503.21875 | -14.286694 | -14.571549 | -14.469669 | -15.037754 | [
"NEUROPLASTIC EXPANSION IN DEEP REINFORCE-\nMENT LEARNING\n\nJiashun Liu\nHKUST\n\nJohan Obando-Ceron\nMila - Qu´ebec AI Institute\nUniversit´e de Montr´eal\n\nAaron Courville\nMila - Qu´ebec AI Institute\nUniversit´e de Montr´eal\n\nLing Pan∗\nHKUST\n\nABSTRACT\n\nThe loss of plasticity in learning agents, analogo... | true |
36 | <insight>
Advancing the capabilities of artificial learning systems requires tackling limitations on multiple fronts, encompassing both the overall network organization and the fundamental mechanisms of its components. One key strategy involves dynamically modifying the network's architecture during training—adding or ... | -2,530.867188 | -2,686.84375 | -2,583.484375 | -2,640.523438 | 5,274.75 | -13.113301 | -13.921471 | -13.385929 | -13.681469 | [
"NEUROPLASTIC EXPANSION IN DEEP REINFORCE-\nMENT LEARNING\n\nJiashun Liu\nHKUST\n\nJohan Obando-Ceron\nMila - Qu´ebec AI Institute\nUniversit´e de Montr´eal\n\nAaron Courville\nMila - Qu´ebec AI Institute\nUniversit´e de Montr´eal\n\nLing Pan∗\nHKUST\n\nABSTRACT\n\nThe loss of plasticity in learning agents, analogo... | true |
36 | <insight>
Addressing fundamental limitations in artificial learning systems, such as declining adaptability over time or difficulty processing complex sequential patterns, can be approached through distinct, biologically-inspired strategies. One path involves dynamically altering the network's overall structure – growi... | -2,665.898438 | -2,407.685547 | -2,467.255859 | -2,788.004883 | 5,394.333008 | -14.256142 | -12.875324 | -13.193882 | -14.909117 | [
"NEUROPLASTIC EXPANSION IN DEEP REINFORCE-\nMENT LEARNING\n\nJiashun Liu\nHKUST\n\nJohan Obando-Ceron\nMila - Qu´ebec AI Institute\nUniversit´e de Montr´eal\n\nAaron Courville\nMila - Qu´ebec AI Institute\nUniversit´e de Montr´eal\n\nLing Pan∗\nHKUST\n\nABSTRACT\n\nThe loss of plasticity in learning agents, analogo... | true |
37 | <insight>
Achieving sophisticated goals with large AI models, such as generating coherent outputs that synthesize multiple distinct learned concepts, often involves complex techniques applied during the model's use (inference). These generation techniques frequently rely on large, computationally demanding underlying m... | -1,785.609375 | -1,910.125 | -1,921.59375 | -1,805.578125 | 3,579.71875 | -11.825228 | -12.649835 | -12.725786 | -11.957471 | [
"TWEEDIEMIX: IMPROVING MULTI-CONCEPT FUSION\nFOR DIFFUSION-BASED IMAGE/VIDEO GENERATION\n\nGihyun Kwon\nKRAFTON\ngkwon@krafton.com\n\nJong Chul Ye\nKim Jaechul Graduate School of AI, KAIST\njong.ye@kaist.ac.kr\n\nFigure 1: Multi-concept Generation Results from TweedieMix. Our model can generate high-\nquality multi... | true |
37 | <insight>
Advancements in generative AI are proceeding along two critical, interdependent paths. One path focuses on enhancing the expressive capability and personalization of models, enabling the generation of sophisticated outputs that integrate multiple distinct concepts or user-specific elements within a single cre... | -2,594.484375 | -2,749 | -2,760.40625 | -2,812.25 | 5,395.328125 | -13.800448 | -14.62234 | -14.683012 | -14.958776 | [
"TWEEDIEMIX: IMPROVING MULTI-CONCEPT FUSION\nFOR DIFFUSION-BASED IMAGE/VIDEO GENERATION\n\nGihyun Kwon\nKRAFTON\ngkwon@krafton.com\n\nJong Chul Ye\nKim Jaechul Graduate School of AI, KAIST\njong.ye@kaist.ac.kr\n\nFigure 1: Multi-concept Generation Results from TweedieMix. Our model can generate high-\nquality multi... | true |
37 | <insight>
Advancements are simultaneously pushing the boundaries of large generative AI models in two key directions: enhancing sophisticated control over the generated output (such as composing multiple distinct, personalized concepts within a single image or video) and improving the efficiency of these massive models... | -1,897.601563 | -2,015.242188 | -2,039.042969 | -1,963.742188 | 3,837.542969 | -12.908854 | -13.70913 | -13.87104 | -13.35879 | [
"TWEEDIEMIX: IMPROVING MULTI-CONCEPT FUSION\nFOR DIFFUSION-BASED IMAGE/VIDEO GENERATION\n\nGihyun Kwon\nKRAFTON\ngkwon@krafton.com\n\nJong Chul Ye\nKim Jaechul Graduate School of AI, KAIST\njong.ye@kaist.ac.kr\n\nFigure 1: Multi-concept Generation Results from TweedieMix. Our model can generate high-\nquality multi... | true |
38 | <insight>
Understanding the theoretical conditions under which systems can generalize learned structural patterns to new scales, such as processing longer sequences than seen during training by leveraging positional information, provides a foundation for what is potentially achievable. However, achieving robust perform... | -2,010.40625 | -1,963.160156 | -1,965.734375 | -2,071.28125 | 4,079.113281 | -12.565039 | -12.269751 | -12.28584 | -12.945508 | [
"A FORMAL FRAMEWORK FOR UNDERSTANDING\nLENGTH GENERALIZATION IN TRANSFORMERS\n\nXinting Huang1∗ Andy Yang2∗ Satwik Bhattamishra3 Yash Sarrof1\nAndreas Krebs4 Hattie Zhou5 Preetum Nakkiran6 Michael Hahn1†\n1Saarland University 2University of Notre Dame\n4University of T¨ubingen 5Mila, Universit´e de Montr´eal\n\n3Un... | true |
38 | <insight>
Theoretical frameworks can establish the potential for certain model architectures to successfully generalize to inputs with structural variations not encountered during training, such as changes in sequence length, under idealized assumptions about the learning process. However, the practical realization of ... | -2,382.3125 | -2,379.445313 | -2,380.640625 | -2,515.195313 | 4,896.3125 | -13.459393 | -13.443193 | -13.449947 | -14.210143 | [
"A FORMAL FRAMEWORK FOR UNDERSTANDING\nLENGTH GENERALIZATION IN TRANSFORMERS\n\nXinting Huang1∗ Andy Yang2∗ Satwik Bhattamishra3 Yash Sarrof1\nAndreas Krebs4 Hattie Zhou5 Preetum Nakkiran6 Michael Hahn1†\n1Saarland University 2University of Notre Dame\n4University of T¨ubingen 5Mila, Universit´e de Montr´eal\n\n3Un... | true |
38 | <insight>
Ensuring models perform reliably on inputs extending beyond their training experience demands distinct approaches based on *how* the new inputs differ. When inputs vary structurally in predictable ways, such as increased sequence length, successful generalization may hinge on the model's inherent capacity to ... | -1,957.328125 | -1,839.5 | -1,841.171875 | -1,901.625 | 3,857.28125 | -12.709923 | -11.944805 | -11.955662 | -12.348214 | [
"A FORMAL FRAMEWORK FOR UNDERSTANDING\nLENGTH GENERALIZATION IN TRANSFORMERS\n\nXinting Huang1∗ Andy Yang2∗ Satwik Bhattamishra3 Yash Sarrof1\nAndreas Krebs4 Hattie Zhou5 Preetum Nakkiran6 Michael Hahn1†\n1Saarland University 2University of Notre Dame\n4University of T¨ubingen 5Mila, Universit´e de Montr´eal\n\n3Un... | true |
39 | <insight>
Advanced fine-tuning approaches for large models are simultaneously deepening our theoretical understanding and expanding practical capabilities. Research is revealing that the effectiveness of certain efficient tuning methods stems from underlying structural principles, such as shared parameterizations impli... | -2,423.78125 | -1,694.523438 | -1,959.78125 | -2,315.960938 | 4,474.484375 | -14.174159 | -9.909493 | -11.460709 | -13.543632 | [
"REVISITING PREFIX-TUNING: STATISTICAL BENEFITS\nOF REPARAMETERIZATION AMONG PROMPTS\n\nMinh Le∗∗1, Chau Nguyen∗1, Huy Nguyen∗2, Quyen Tran1, Trung Le3, Nhat Ho2\n1 Movian AI, Vietnam 2 The University of Texas at Austin\n\n3 Monash University\n\nABSTRACT\n\nPrompt-based techniques, such as prompt-tuning and prefix-... | true |
39 | <insight>
Fine-tuning large models offers a powerful way to embed sophisticated properties directly into their parameters, going beyond simple task adaptation. Specific parameterization strategies employed during efficient fine-tuning can implicitly create beneficial shared structures that enhance learning efficiency. ... | -1,264.523438 | -1,393.84375 | -1,391.578125 | -1,267.984375 | 2,534.773438 | -10.0359 | -11.062252 | -11.044271 | -10.063368 | [
"REVISITING PREFIX-TUNING: STATISTICAL BENEFITS\nOF REPARAMETERIZATION AMONG PROMPTS\n\nMinh Le∗∗1, Chau Nguyen∗1, Huy Nguyen∗2, Quyen Tran1, Trung Le3, Nhat Ho2\n1 Movian AI, Vietnam 2 The University of Texas at Austin\n\n3 Monash University\n\nABSTRACT\n\nPrompt-based techniques, such as prompt-tuning and prefix-... | true |
39 | <insight>
Advanced fine-tuning techniques for large models can enhance performance and efficiency through fundamentally different approaches. One approach reveals that specific parameterization strategies, like those used in some prompt-based methods, implicitly encode beneficial structural constraints (e.g., shared st... | -2,222.359375 | -2,335.234375 | -2,323.046875 | -2,328.203125 | 4,562.75 | -12.699197 | -13.344196 | -13.274553 | -13.304018 | [
"REVISITING PREFIX-TUNING: STATISTICAL BENEFITS\nOF REPARAMETERIZATION AMONG PROMPTS\n\nMinh Le∗∗1, Chau Nguyen∗1, Huy Nguyen∗2, Quyen Tran1, Trung Le3, Nhat Ho2\n1 Movian AI, Vietnam 2 The University of Texas at Austin\n\n3 Monash University\n\nABSTRACT\n\nPrompt-based techniques, such as prompt-tuning and prefix-... | true |
41 | <insight>
Developing sophisticated intelligent systems, whether for interacting with the physical world based on instructions or for modeling complex physical phenomena, often encounters limitations with rigid, multi-stage processing or fixed solution algorithms. A powerful alternative involves designing the core syste... | -2,414.65625 | -2,443.65625 | -2,469.421875 | -2,417.96875 | 4,806.859375 | -13.798036 | -13.96375 | -14.110982 | -13.816964 | [
"VLAS: VISION-LANGUAGE-ACTION MODEL WITH\nSPEECH INSTRUCTIONS FOR CUSTOMIZED ROBOT\nMANIPULATION\n\nWei Zhao1 Pengxiang Ding1,2 Min Zhang1 Zhefei Gong1 Shuanghao Bai3\nHan Zhao1,2 Donglin Wang1∗\n1Westlake University 2Zhejiang University 3Xi’an Jiaotong University\n\nABSTRACT\n\nVision-language-action models (VLAs)... | true |
41 | <insight>
Advanced machine learning systems are increasingly being designed to handle instance-specific variations, moving beyond generic processing. One facet of this involves enabling physical systems, like robots, to interpret nuanced, multimodal user commands, including non-semantic aspects of speech such as voice ... | -2,726.40625 | -2,724.5 | -2,804.84375 | -2,767.0625 | 5,413.125 | -14.42543 | -14.415344 | -14.840443 | -14.640542 | [
"VLAS: VISION-LANGUAGE-ACTION MODEL WITH\nSPEECH INSTRUCTIONS FOR CUSTOMIZED ROBOT\nMANIPULATION\n\nWei Zhao1 Pengxiang Ding1,2 Min Zhang1 Zhefei Gong1 Shuanghao Bai3\nHan Zhao1,2 Donglin Wang1∗\n1Westlake University 2Zhejiang University 3Xi’an Jiaotong University\n\nABSTRACT\n\nVision-language-action models (VLAs)... | true |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.