title
stringlengths
6
157
paper_url
stringlengths
42
42
authors
listlengths
1
84
type
stringclasses
2 values
primary_area
stringclasses
21 values
abstract
large_stringlengths
424
2.66k
keywords
listlengths
1
42
TL;DR
large_stringlengths
6
250
submission_number
int64
3
25.6k
arxiv_id
stringlengths
10
10
arxiv_id_source
stringclasses
2 values
embedding
listlengths
768
768
Task Tokens: A Flexible Approach to Adapting Behavior Foundation Models
https://openreview.net/forum?id=6T3wJQhvc3
[ "Ron Vainshtein", "Zohar Rimon", "Shie Mannor", "Chen Tessler" ]
Poster
reinforcement learning
Recent advancements in imitation learning for robotic control have led to transformer-based behavior foundation models (BFMs) that enable multi-modal, human-like control for humanoid agents. These models generate solutions when conditioned on high-level goals or prompts, for example, walking to a coordinate when condit...
[ "Reinforcement Learning", "Hierarchial Reinforcement Learning", "Behavior Foundation Models", "Humanoid Control" ]
Task Tokens enable task-specific adaptation of behavior foundation models by learning a reinforcement-trained encoder, enhancing control without compromising generalization.
25,607
2503.22886
title_snapshot
[ -0.0008359543862752616, -0.030916187912225723, -0.015673547983169556, 0.014925187453627586, 0.035812027752399445, 0.03931316360831261, 0.021506335586309433, 0.03143975883722305, -0.03957877680659294, -0.040012091398239136, -0.021476345136761665, 0.01585039496421814, -0.055820003151893616, ...
Decomposing LLM Computation with Jets
https://openreview.net/forum?id=u6JLh0BO5h
[ "Yihong Chen", "Xiangxiang Xu", "Pontus Stenetorp", "Sebastian Riedel", "Luca Franceschi" ]
Poster
interpretability and explainable AI
Large language models are becoming general knowledge engines for diverse applications. However, their computations are deeply entangled after training, resisting modularization which complicates interpretability, auditing, and long-term maintenance. We introduce Jet Expansions, a framework for expanding computational g...
[ "decomposition", "transformer", "neural-symbolic", "n-grams", "interpretability", "controllability" ]
We introduce jet expansions: operators that "cuts through" LLM entaglement, separating parts of computation of interest and enabling systematic model inspection like n-gram tables
25,587
null
null
[ -0.025453023612499237, -0.002427687868475914, -0.010450122877955437, 0.02993519976735115, 0.03797256946563721, 0.036146216094493866, 0.03750168904662132, -0.012011468410491943, -0.012975707650184631, -0.008761564269661903, -0.014399932697415352, 0.043763112276792526, -0.07323324680328369, ...
Neuron-Aware Data Selection in Instruction Tuning for Large Language Models
https://openreview.net/forum?id=uq6UWRgzMr
[ "Xin Chen", "Junchao Wu", "Shu Yang", "Runzhe Zhan", "Zeyu Wu", "Min Yang", "Shujian Huang", "Lidia S. Chao", "Derek F. Wong" ]
Poster
interpretability and explainable AI
Instruction Tuning (IT) has been proven to be an effective approach to unlock the powerful capabilities of large language models (LLMs). Recent studies indicate that excessive IT data can degrade LLMs performance, while carefully selecting a small subset of high-quality IT data can significantly enhance their capabili...
[ "Instruction Tuning", "Data Selection", "Large Language Models" ]
NAIT is an efficient algorithm that selects high-quality instruction tuning data by analyzing neuron activation pattern similarity, enhancing large language models' performance and general capabilities.
25,580
2603.13201
title_snapshot
[ -0.04062799736857414, -0.03215198218822479, -0.018052879720926285, 0.035482555627822876, 0.05305304750800133, 0.03716743364930153, 0.02434348315000534, 0.009067131206393242, -0.02186923287808895, -0.007715132553130388, -0.030404694378376007, 0.03146608546376228, -0.03973812609910965, 0.001...
Quadratic Direct Forecast for Training Multi-Step Time-Series Forecast Models
https://openreview.net/forum?id=vpO8n9AqEG
[ "Hao Wang", "Licheng Pan", "Yuan Lu", "Zi Ciu Chan", "Tianqiao Liu", "Shuting He", "Zhixuan Chu", "Qingsong Wen", "Haoxuan Li", "Zhouchen Lin" ]
Poster
learning on time series and dynamical systems
The design of training objective is central to training time-series forecasting models. Existing training objectives such as mean squared error mostly treat each future step as an independent, equally weighted task, which we found leading to the following two issues: (1) overlook the *label autocorrelation effect* amon...
[ "Time-series", "time-series forecast" ]
null
25,573
2511.00053
title_snapshot
[ -0.03691662475466728, -0.037040743976831436, 0.01939641498029232, 0.025249414145946503, 0.025342818349599838, 0.037225354462862015, 0.010771355591714382, -0.011487883515655994, -0.042199041694402695, -0.04472785443067551, -0.033877480775117874, 0.020247995853424072, -0.06369523704051971, 0...
Submodular Function Minimization with Dueling Oracle
https://openreview.net/forum?id=BeMtzSH1d7
[ "Kaien Sho", "Shinji Ito" ]
Poster
optimization
We consider submodular function minimization using a *dueling oracle*, a noisy pairwise comparison oracle that provides relative feedback on function values between two queried sets. The oracle's responses are governed by a *transfer function*, which characterizes the relationship between differences in function values...
[ "submodular minimization", "deling oracle", "preference-based optimization" ]
We study submodular minimization with a dueling oracle giving noisy pairwise feedback.
25,553
null
null
[ -0.043026674538850784, 0.003168470924720168, -0.024399466812610626, 0.055632852017879486, 0.034934304654598236, 0.05185055360198021, 0.006822189781814814, -0.012955239042639732, -0.010849904268980026, -0.021143266931176186, 0.005432602949440479, 0.03393084555864334, -0.07723353058099747, -...
Improving Semantic Proximity in Information Retrieval through Cross-Lingual Alignment
https://openreview.net/forum?id=NvKvW5k6Kk
[ "Seongtae Hong", "Youngjoon Jang", "Jungseob Lee", "Hyeonseok Moon", "Heuiseok Lim" ]
Poster
unsupervised, self-supervised, semi-supervised, and supervised representation learning
With the increasing accessibility and utilization of multilingual documents, Cross-Lingual Information Retrieval (CLIR) has emerged as an important research area. Conventionally, CLIR tasks have been conducted under settings where the language of documents differs from that of queries, and typically, the documents are ...
[ "Cross-Lingual Alignment", "Information Retrieval", "Multilingual Embedding", "Cross-Lingual Information Retrieval" ]
This paper identifies multilingual embedding gaps in cross-lingual retrieval, proposes scenario and Max@R metric, and introduces a training strategy combining JSD and InfoNCE loss, significantly improving cross-lingual alignment with minimal data.
25,552
2604.05684
title_snapshot
[ -0.006696826778352261, -0.030422106385231018, 0.0033992365933954716, 0.05462192744016647, 0.020853208377957344, -0.03778091445565224, 0.00326559878885746, 0.01831672713160515, 0.0030863075517117977, -0.01499023474752903, -0.05078410357236862, 0.011413982138037682, -0.05697387084364891, 0.0...
Test-Time Accuracy-Cost Control in Neural Simulators via Recurrent-Depth
https://openreview.net/forum?id=U2j9ZNgHqw
[ "Harris Abdul Majid", "Pietro Sittoni", "Francesco Tudisco" ]
Poster
applications to physical sciences (physics, chemistry, biology, etc.)
Accuracy-cost trade-offs are a fundamental aspect of scientific computing. Classical numerical methods inherently offer such a trade-off: increasing resolution, order, or precision typically yields more accurate solutions at higher computational cost. We introduce \textbf{Recurrent-Depth Simulator} (\textbf{RecurrSim})...
[ "Neural Simulator", "Recurrent Depth", "AI4Simulation" ]
null
25,526
null
null
[ -0.04043664038181305, 0.00837439950555563, 0.014256698079407215, 0.042583271861076355, 0.05038338899612427, 0.05198339372873306, 0.033738814294338226, 0.012487906031310558, -0.009086400270462036, -0.05405976623296738, 0.01964527554810047, -0.0046523865312337875, -0.037042856216430664, 0.01...
CORDS - Continuous Representations of Discrete Structures
https://openreview.net/forum?id=RObkOKADBU
[ "Tin Hadži Veljković", "Erik J Bekkers", "Michael Tiemann", "Jan-Willem van de Meent" ]
Poster
learning on graphs and other geometries & topologies
Many learning problems require predicting sets of objects when the number of objects is not known beforehand. Examples include object detection, molecular modeling, and scientific inference tasks such as astrophysical source detection. Existing methods often rely on padded representations or must explicitly infer the s...
[ "Continuous set representations", "Neural fields", "Variable-cardinality prediction", "Invertible encoding/decoding", "Diffusion and flow matching", "Object detection", "Molecular generation", "Simulation-based inference" ]
We turn discrete objects into continuous fields that implicitly encode their count, offering a simple way to handle variable cardinality across tasks and domains.
25,519
2601.21583
title_snapshot
[ -0.009147780947387218, 0.000011557815923879389, -0.021373383700847626, 0.04003588855266571, 0.0395980030298233, 0.02266901731491089, 0.0005451223696582019, 0.0018911975203081965, -0.029355177655816078, -0.043903883546590805, -0.0342218391597271, 0.007476929109543562, -0.08012612909078598, ...
MedAraBench: Large-scale Arabic Medical Question Answering Dataset and Benchmark
https://openreview.net/forum?id=1BXojAgNrg
[ "Mouath Abu Daoud", "Leen Kharouf", "Omar El Hajj", "Dana El Samad", "Mariam Al-Omari", "Jihad Mallat", "Khaled Saleh", "Nizar Habash", "Farah E. Shamout" ]
Poster
datasets and benchmarks
Arabic remains one of the most underrepresented languages in natural language processing research, particularly in medical applications, due to the limited availability of open-source data and benchmarks. The lack of resources hinders efforts to evaluate and advance the multilingual capabilities of Large Language Model...
[ "Dataset Benchmark", "Large Language Models", "Arabic Natural Language Processing", "Medical Question Answering" ]
null
25,508
2602.01714
title_snapshot
[ -0.01637825183570385, -0.026904914528131485, -0.0386020690202713, 0.024014629423618317, 0.023853477090597153, 0.04193916544318199, 0.019984696060419083, 0.002992071444168687, -0.002412965055555105, -0.024825429543852806, -0.03215980529785156, 0.05043744668364525, -0.08010181039571762, -0.0...
Fracture-GS: Dynamic Fracture Simulation with Physics-Integrated Gaussian Splatting
https://openreview.net/forum?id=zcAwK50ft0
[ "Xiaogang Wang", "Hongyu Wu", "Wenfeng Song", "Kai Xu" ]
Poster
applications to robotics, autonomy, planning
This paper presents a unified framework for simulating and visualizing dynamic fracture phenomena in extreme mechanical collisions using multi-view image inputs. While existing methods primarily address elastic deformations at contact surfaces, they fail to capture the complex physics of extreme collisions, often produ...
[ "3D vision", "Physics-based Simulation" ]
null
25,504
null
null
[ -0.009710283018648624, 0.01195431500673294, -0.007936838082969189, 0.028285209089517593, 0.02693546935915947, 0.04063718020915985, 0.02437649480998516, 0.005024835932999849, -0.06427113711833954, -0.06831810623407364, -0.006096027325838804, -0.0007057127659209073, -0.03342131897807121, 0.0...
High Probability Bounds for Non-Convex Stochastic Optimization with Momentum
https://openreview.net/forum?id=KirKWFPYJA
[ "Shaojie Li", "Pengwei Tang", "Bowei Zhu", "Yong Liu" ]
Poster
learning theory
Stochastic gradient descent with momentum (SGDM) is widely used in machine learning, yet high-probability learning bounds for SGDM in non-convex settings remain scarce. In this paper, we provide high-probability convergence bounds and generalization bounds for SGDM. First, we establish such bounds for the gradient norm...
[ "Momentum", "nonconvex learning", "generalization" ]
null
25,501
null
null
[ -0.042766522616147995, -0.011598692275583744, 0.012426957488059998, 0.04097757115960121, 0.02617339976131916, 0.02721676044166088, 0.029058974236249924, -0.009636279195547104, -0.016881030052900314, -0.03756043314933777, -0.011967272497713566, -0.02287435717880726, -0.05028625205159187, -0...
PoinnCARE: Hyperbolic Multi-Modal Learning for Enzyme Classification
https://openreview.net/forum?id=dGxAYNK6JU
[ "Kun XIE", "Peng Zhou", "Xingyi Zhang", "Wei Liu", "Peilin Zhao", "Sibo Wang", "Biaobin Jiang" ]
Poster
applications to physical sciences (physics, chemistry, biology, etc.)
Enzyme Commission (EC) number prediction is vital for elucidating enzyme functions and advancing biotechnology applications. However, current methods struggle to capture the hierarchical relationships among enzymes and often overlook critical structural and active site features. To bridge this gap, we introduce PoinnCA...
[ "EC number prediction", "enzyme function", "hyperbolic space learning", "multi-modal learning", "enzyme structure", "enzyme active site" ]
null
25,500
null
null
[ -0.019870338961482048, -0.04766073822975159, -0.01873438060283661, 0.04739838466048241, 0.029690830036997795, 0.00947307888418436, 0.0319969579577446, -0.014134720899164677, 0.007183796726167202, -0.04110254347324371, 0.029802897945046425, 0.012932807207107544, -0.05880551040172577, 0.0007...
FingerTip 20K: A Benchmark for Proactive and Personalized Mobile LLM Agents
https://openreview.net/forum?id=n3iFV0gLMc
[ "Qinglong Yang", "Haoming Li", "Haotian Zhao", "Xiaokai Yan", "Jingtao Ding", "Fengli Xu", "Yong Li" ]
Poster
datasets and benchmarks
Mobile GUI agents are becoming critical tools to improve user experience on smart devices, with multimodal large language models (MLLMs) emerging as the dominant paradigms in this domain. Current agents, however, rely on explicit human instructions, overlooking the potential to leverage the contextual information (like...
[ "Mobile Agent", "LLM Agent", "GUI", "Proactive Agent", "Personalization" ]
null
25,468
2507.21071
title_snapshot
[ -0.023786868900060654, -0.028025228530168533, 0.029995471239089966, 0.0049143703654408455, 0.05791344866156578, -0.012289809994399548, 0.03430687636137009, 0.010650706477463245, -0.011042761616408825, -0.020353442057967186, 0.0023095363285392523, 0.014592391438782215, -0.073185034096241, -...
Characterizing and Optimizing the Spatial Kernel of Multi Resolution Hash Encodings
https://openreview.net/forum?id=q05hC1Pzkr
[ "Tianxiang Dai", "Jonathan Fan" ]
Poster
applications to computer vision, audio, language, and other modalities
Multi-Resolution Hash Encoding (MHE), the foundational technique behind Instant Neural Graphics Primitives, provides a powerful parameterization for neural fields. However, its spatial behavior lacks rigorous understanding from a physical systems perspective, leading to reliance on heuristics for hyperparameter selecti...
[ "multi-resolution hash encoding", "implicit neural representations", "neural fields", "point spread function", "spatial kernel analysis", "anisotropy", "resolution limit", "FWHM", "hash collisions", "signal-to-noise ratio", "NeRF" ]
We analyze Multi-Resolution Hash Encoding (MHE) using its Point Spread Function (PSF) to reveal that effective resolution is governed by average, not finest, resolution, and introduce Rotated MHE to mitigate inherent anisotropy and collision noise.
25,450
2602.10495
title_snapshot
[ -0.039382193237543106, 0.032644566148519516, 0.011089514940977097, 0.047400761395692825, 0.03819405660033226, 0.04162555932998657, 0.001062441268004477, -0.010292566381394863, -0.04898132011294365, -0.05427838861942291, 0.008405832573771477, -0.027717549353837967, -0.03135077282786369, -0....
EVEREST: A Transformer for Probabilistic Rare-Event Anomaly Detection with Evidential and Tail-Aware Uncertainty
https://openreview.net/forum?id=ScpCaOVGw1
[ "Antanas Žilinskas", "Robert Noel Shorten", "Jakub Marecek" ]
Poster
learning on time series and dynamical systems
Forecasting rare events in multivariate time-series data is a central challenge in machine learning, complicated by severe class imbalance, long-range dependencies, and distributional uncertainty. We introduce EVEREST, a transformer-based architecture for probabilistic rare-event forecasting that delivers calibrated pr...
[ "Transformer models", "Uncertainty quantification", "Evidential deep learning", "Extreme value theory", "Imbalanced classification" ]
EVEREST is a transformer architecture for rare-event time-series forecasting that combines evidential and tail-aware uncertainty to deliver calibrated, interpretable, and state-of-the-art predictions across scientific anomaly detection tasks.
25,448
2601.19022
title_judge
[ 0.001887625316157937, -0.042065318673849106, -0.0010761877056211233, 0.03137300908565521, 0.04945498704910278, -0.0045815203338861465, 0.030580928549170494, -0.004361292812973261, -0.06514891982078552, -0.019800325855612755, -0.03766573593020439, 0.008084913715720177, -0.07071757316589355, ...
How Well Does GPT-4o Understand Vision? Evaluating Multimodal Foundation Models on Standard Computer Vision Tasks
https://openreview.net/forum?id=Oq3yRhFp0t
[ "Rahul Ramachandran", "Ali Garjani", "Roman Bachmann", "Andrei Atanov", "Oğuzhan Fatih Kar", "Amir Zamir" ]
Poster
unsupervised, self-supervised, semi-supervised, and supervised representation learning
Multimodal foundation models, such as GPT-4o, have recently made remarkable progress, but it is not clear where exactly these models stand in terms of understanding vision. In this paper, we benchmark the performance of popular multimodal foundation models (GPT-4o, o4-mini, Gemini 1.5 Pro and Gemini 2.0 Flash, Claude 3...
[ "vision benchmark", "multimodal foundation models", "vision language models", "standard computer vision tasks" ]
null
25,423
2507.01955
title_snapshot
[ 0.01134119089692831, -0.03568432107567787, 0.020530277863144875, 0.04291063919663429, 0.02916247956454754, 0.025161851197481155, 0.002411199500784278, 0.05136023461818695, -0.0418788343667984, -0.03331391140818596, -0.029936542734503746, 0.040217529982328415, -0.08126510679721832, -0.00179...
Common Corpus: The Largest Collection of Ethical Data for LLM Pre-Training
https://openreview.net/forum?id=0wSlFpMsGb
[ "Pierre-Carl Langlais", "Pavel Chizhov", "Catherine Arnett", "Carlos Rosas Hinostroza", "Mattia Nee", "Eliot Krzysztof Jones", "Irène Girard", "David Mach", "Anastasia Stasenko", "Ivan P. Yamshchikov" ]
Oral
datasets and benchmarks
Large Language Models (LLMs) are pre-trained on large amounts of data from different sources and domains. Such datasets often contain trillions of tokens, including large portions of copyrighted or proprietary content, which raises questions about the legal use of such models. This underscores the need for truly open p...
[ "dataset", "pre-training", "large language models", "open data", "open science", "multilingual" ]
We assemble and release the largest truly open multilingual dataset for LLM pre-training consisting of 2 trillion tokens
25,369
2506.01732
title_snapshot
[ -0.012947111390531063, -0.05378737300634384, -0.024764589965343475, 0.02979375049471855, 0.0672144815325737, -0.003873468143865466, 0.006588546093553305, 0.026918848976492882, -0.02197900600731373, 0.01625223085284233, -0.053208060562610626, 0.008601650595664978, -0.07236508280038834, -0.0...
Unlocking the Potential of Weighting Methods in Federated Learning Through Communication Compression
https://openreview.net/forum?id=zOWljZMbCm
[ "Valery Parfenov", "Nail Bashirov", "Daniil Medyakov", "Dmitry Bylinkin", "Aleksandr Beznosikov" ]
Poster
optimization
Modern machine learning problems are frequently formulated in federated learning domain and incorporate inherently heterogeneous data. Weighting methods operate efficiently in terms of iteration complexity and represent a common direction in this setting. At the same time, they do not address directly the main obstacle...
[ "Convex optimization", "Compression", "Stochastic optimization" ]
null
25,365
null
null
[ -0.007463356014341116, -0.054443225264549255, 0.0047484529204666615, 0.046391621232032776, 0.03529611974954605, 0.05259128287434578, -0.0017357026226818562, -0.022495808079838753, -0.006960927043110132, -0.0588686466217041, -0.026549888774752617, -0.014691539108753204, -0.05537014454603195, ...
Masked Skill Token Training for Hierarchical Off-Dynamics Transfer
https://openreview.net/forum?id=K4ngUOra9m
[ "Zeyu Feng", "Haiyan Yin", "Yew-Soon Ong", "Harold Soh" ]
Poster
reinforcement learning
Generalizing policies across environments with altered dynamics remains a key challenge in reinforcement learning, particularly in offline settings where direct interaction or fine-tuning is impractical. We introduce Masked Skill Token Training (MSTT), a fully offline hierarchical RL framework that enables policy trans...
[ "Tranfser Learning", "Skills", "Hierarchical RL", "Embodied AI" ]
null
25,348
null
null
[ -0.020601820200681686, -0.022299857810139656, -0.008987263776361942, 0.04335622489452362, 0.037181556224823, 0.024630585685372353, 0.04399072751402855, -0.008202292025089264, -0.06141937896609306, -0.036042727530002594, -0.03365437686443329, 0.0015092437388375401, -0.06342650204896927, -0....
Spilling the Beans: Teaching LLMs to Self-Report Their Hidden Objectives
https://openreview.net/forum?id=sWs0cCuM8I
[ "Chloe Li", "Mary Phuong", "Daniel Tan" ]
Poster
alignment, fairness, safety, privacy, and societal considerations
As AI systems become more capable of complex agentic tasks, they also become more capable of pursuing undesirable objectives and causing harm. Previous work has attempted to catch these unsafe instances by interrogating models directly about their objectives and behaviors. However, the main weakness of trusting interro...
[ "honesty", "honesty finetuning", "interrogation", "alignment auditing" ]
We propose a SFT method that trains models to admit simple factual errors, which generalizes to admitting hidden objectives in sabotage tasks under adversarial pressure to conceal them, improving techniques for incriminating misaligned AI systems.
25,344
2511.06626
title_snapshot
[ -0.0012156565207988024, -0.007974356412887573, 0.006042103748768568, 0.012212153524160385, 0.051993511617183685, 0.006903639528900385, 0.040107615292072296, -0.004533963743597269, -0.030754543840885162, -0.02096952311694622, -0.027526792138814926, 0.05182329937815666, -0.06119413301348686, ...
Fairness via Independence: A General Regularization Framework for Machine Learning
https://openreview.net/forum?id=sbEb0Ld6MK
[ "Yezi Liu", "Hanning Chen", "Wenjun Huang", "Yang Ni", "Mohsen Imani" ]
Poster
alignment, fairness, safety, privacy, and societal considerations
Fairness in machine learning has emerged as a central concern, as predictive models frequently inherit or even amplify biases present in training data. Such biases often manifest as unintended correlations between model outcomes and sensitive attributes, leading to systematic disparities across demographic groups. Exis...
[ "Bias Mitigation", "Statistical Independence", "Fairness in Machine Learning" ]
We introduce a general framework to promote fairness in machine learning by reducing the dependence between model predictions and sensitive attributes.
25,320
null
null
[ -0.01073729619383812, -0.019950570538640022, -0.011326897889375687, 0.020928794518113136, 0.04358725994825363, 0.023680442944169044, 0.024374576285481453, -0.00676223449409008, -0.06317415833473206, -0.032052185386419296, 0.003587809856981039, 0.008655247278511524, -0.07986408472061157, -0...
Transformers as Unsupervised Learning Algorithms: A study on Gaussian Mixtures
https://openreview.net/forum?id=4hKNGmjXVQ
[ "Zhiheng Chen", "Ruofan Wu", "Guanhua Fang" ]
Poster
learning theory
The transformer architecture has demonstrated remarkable capabilities in modern artificial intelligence, among which the capability of implicitly learning an internal model during inference time is widely believed to play a key role in the understanding of pre-trained large language models. However, most recent works h...
[ "In-context learning", "Gaussian Mixture Models", "Theory" ]
null
25,315
2505.11918
title_snapshot
[ -0.003944553900510073, -0.03674423322081566, -0.010730602778494358, 0.027141442522406578, 0.007924959994852543, 0.010919156484305859, 0.01543356478214264, 0.02332010120153427, -0.02353346347808838, -0.0073034316301345825, -0.027183009311556816, 0.05986879765987396, -0.06313760578632355, 0....
Q-RAG: Long Context Multi‑Step Retrieval via Value‑Based Embedder Training
https://openreview.net/forum?id=MS9nWFY7LG
[ "Artyom Sorokin", "Nazar Buzun", "Aleksandr Anokhin", "Egor KONSTANTINOVICH VEDERNIKOV", "Petr Anokhin", "Mikhail Burtsev", "Evgeny Burnaev" ]
Oral
reinforcement learning
Retrieval-Augmented Generation (RAG) methods enhance LLM performance by efficiently filtering relevant context for LLMs, reducing hallucinations and inference cost. However, most existing RAG methods focus on single-step retrieval, which is often insufficient for answering complex questions that require multi-step sear...
[ "Reinforcement Learning", "RL", "QA", "Long-context", "RAG", "NLP" ]
null
25,302
2511.07328
title_snapshot
[ -0.0010835888097062707, -0.04703739285469055, -0.011473975144326687, 0.028993776068091393, 0.031679049134254456, -0.028867391869425774, 0.006095395889133215, -0.019456597045063972, -0.025735966861248016, -0.012465998530387878, -0.05942172929644585, 0.03738884627819061, -0.08094984292984009, ...
THEMIS: Towards Holistic Evaluation of MLLMs for Scientific Paper Fraud Forensics
https://openreview.net/forum?id=y3UkklvoW9
[ "Tzu-Yen Ma", "Bo Zhang", "Zichen Tang", "Junpeng Ding", "Haolin Tian", "Yuanze Li", "Zhuodi Hao", "Zixin Ding", "Zirui Wang", "Xinyu Yu", "Shiyao Peng", "Yizhuo Zhao", "Ruomeng Jiang", "Yiling Huang", "Peizhi Zhao", "Jiayuan Chen", "Weisheng Tan", "Haocheng Gao", "Yang Liu", "...
Poster
datasets and benchmarks
We present **THEMIS**, a novel multi-task benchmark designed to comprehensively evaluate multimodal large language models (MLLMs) on visual fraud reasoning within real-world academic scenarios. Compared to existing benchmarks, THEMIS introduces three major advances. (1) **Real-World Scenarios and Complexity**: Our benc...
[ "Multimodal Large Language Model", "Vision Fraud Reasoning", "Scientific Paper Fraud Detection", "Benchmark" ]
We present THEMIS, a holistic multi-task benchmark of over 4000 questions derived from authentic retracted-paper cases and realistically simulated synthetic data, to systematically evaluate the fine-grained visual fraud reasoning abilities of MLLMs.
25,299
2603.25089
title_snapshot
[ 0.006894868332892656, -0.020040789619088173, 0.026700982823967934, 0.059171054512262344, 0.04374997690320015, 0.0017386619001626968, 0.009602369740605354, 0.01519683189690113, -0.02458541840314865, -0.0010357749415561557, -0.0009536428260616958, 0.036689285188913345, -0.05621883645653725, ...
Towards Text-Mask Consistency in Medical Image Segmentation
https://openreview.net/forum?id=riOevy2RwZ
[ "Jie Gui", "HangTu", "Wen Sha", "Xiuquan Du" ]
Poster
other topics in machine learning (i.e., none of the above)
Vision-language models for medical image segmentation often produce masks that conflict with the accompanying text, especially under multi-site/multi-lesion descriptions. We trace this failure to two factors: (i) highly templated and repetitive clinical language causes one-to-one hard contrastive learning to yield nume...
[ "Medical image segmentation", "Vision language models", "Multimodal learning", "Kolmogorov–Arnold Networks" ]
null
25,292
null
null
[ 0.014517327770590782, -0.0005650055245496333, -0.02590278349816799, 0.03665575012564659, 0.01230046059936285, 0.03276842087507248, 0.040524765849113464, 0.03898381069302559, -0.03914297744631767, -0.05943720042705536, -0.038863688707351685, 0.032386213541030884, -0.04445278272032738, 0.005...
SCRAPL: Scattering Transform with Random Paths for Machine Learning
https://openreview.net/forum?id=RuYwbd5xYa
[ "Christopher Mitcheltree", "Vincent Lostanlen", "Emmanouil Benetos", "Mathieu Lagrange" ]
Poster
learning on time series and dynamical systems
The Euclidean distance between wavelet scattering transform coefficients (known as paths) provides informative gradients for perceptual quality assessment of deep inverse problems in computer vision, speech, and audio processing. However, these transforms are computationally expensive when employed as differentiable lo...
[ "scattering transform", "wavelets", "stochastic optimization", "ddsp", "perceptual quality assessment" ]
A stochastic optimization scheme for efficient perceptual quality assessment of deep inverse problems, implemented for differentiable joint time–frequency scattering, with applications to unsupervised sound matching of the Roland TR-808 drum machine.
25,282
2602.11145
title_snapshot
[ 0.007557682227343321, 0.000775298336520791, -0.0237746499478817, 0.022335108369588852, 0.03795546665787697, 0.034208592027425766, 0.020985886454582214, 0.02474314346909523, -0.05577879771590233, -0.07854044437408447, -0.035374321043491364, 0.018973765894770622, -0.04254807531833649, -0.024...
From Text to Talk: Audio-Language Model Needs Non-Autoregressive Joint Training
https://openreview.net/forum?id=e3XLWHFrnr
[ "Tianqiao Liu", "Xueyi Li", "Hao Wang", "Haoxuan Li", "Zhichao Chen", "Weiqi Luo", "Zitao Liu" ]
Poster
foundation or frontier models, including LLMs
Recent advances in large language models (LLMs) have attracted significant interest in extending their capabilities to multimodal scenarios, particularly for speech-to-speech (S2S) conversational systems. However, existing multimodal models handling interleaved audio and text rely on autoregressive (AR) methods, overlo...
[ "Large Multimodal Models", "Multi-token Prediction", "Non-Autoregressive Learning" ]
null
25,264
2509.20072
title_snapshot
[ 0.002431832253932953, -0.022732378914952278, -0.0066496445797383785, 0.046563904732465744, 0.013983016833662987, 0.03458193689584732, 0.06281907111406326, 0.008723694831132889, -0.025390230119228363, -0.033257968723773956, -0.02413863316178322, 0.06855396926403046, -0.0498817153275013, -0....
Not Search, But Scan: Benchmarking MLLMs on Scan-Oriented Academic Paper Reasoning
https://openreview.net/forum?id=GDA1yB6yDP
[ "Rongjin Li", "Zichen Tang", "Xianghe Wang", "Xinyi Hu", "Zhengyu Wang", "Zhengyu Lu", "Yiling Huang", "Jiayuan Chen", "Weisheng Tan", "Jiacheng Liu", "Zhongjun Yang", "Haihong E" ]
Poster
datasets and benchmarks
With the rapid progress of multimodal large language models (MLLMs), AI already performs well at literature retrieval and certain reasoning tasks, serving as a capable assistant to human researchers, yet it remains far from autonomous research. The fundamental reason is that current work on academic paper reasoning is ...
[ "Multimodal Large Language Models", "Academic Paper Reasoning", "Scan-Oriented Reasoning" ]
We present ScholScan, a scan-oriented benchmark for full-paper scholarly reasoning that requires models to build a paper-level evidence view; spanning 1,800 questions from 715 papers, which exposes MLLM gaps and shows RAG ineffective.
25,245
2603.28651
title_snapshot
[ -0.022581635043025017, -0.016278253868222237, 0.0043747988529503345, 0.03996874392032623, 0.025316113606095314, -0.02974710240960121, 0.028933443129062653, 0.016610296443104744, -0.0317453071475029, 0.004931765608489513, -0.019647056236863136, 0.024169625714421272, -0.04167459160089493, -0...
FrugalRAG: Less is More in RL Finetuning for Multi-hop Question Answering
https://openreview.net/forum?id=uQKtwdJN0o
[ "Abhinav Java", "Srivathsan Koundinyan", "Nagarajan Natarajan", "Amit Sharma" ]
Poster
foundation or frontier models, including LLMs
Reinforcement learning (RL) based on the final answer's reward has driven recent progress in small language models (SLMs) on reasoning-heavy tasks such as math and code. However, applying the same techniques to retrieval-augmented generation (RAG) benchmarks like multi-hop QA has yielded limited gains—often trailing su...
[ "Multi-Hop RAG", "Efficiency", "Reasoning", "SLMs" ]
null
25,237
2507.07634
title_snapshot
[ -0.017406301572918892, -0.03901084512472153, 0.0225948728621006, 0.07759351283311844, 0.04683690145611763, 0.025408167392015457, 0.04150674492120743, -0.003877941519021988, -0.04563548415899277, -0.004149705171585083, -0.035563912242650986, 0.0449601411819458, -0.06599545478820801, -0.0492...
Any-step Generation via N-th Order Recursive Consistent Velocity Field Estimation
https://openreview.net/forum?id=GnawtLKGkP
[ "Peng Sun", "Tao Lin" ]
Poster
generative models
Recent advances in few-step generative models (typically $1$-$8$ steps), such as consistency models, have yielded impressive performance. However, their broader adoption is hindered by significant challenges, including substantial computational overhead, the reliance on complex multi-component loss functions, and intri...
[ "Generative Models" ]
null
25,236
null
null
[ -0.02324218861758709, -0.032452840358018875, 0.025406725704669952, 0.04318954795598984, 0.023600751534104347, 0.013223041780292988, 0.038031209260225296, 0.014548989944159985, -0.02339986339211464, -0.0604570135474205, 0.04196805879473686, -0.011543571949005127, -0.07123211771249771, 0.008...
From Broad Exploration to Stable Synthesis: Entropy-Guided Optimization for Autoregressive Image Generation
https://openreview.net/forum?id=NCLjpR2MDq
[ "Han Song", "Yucheng Zhou", "Jianbing Shen", "Yu Cheng" ]
Poster
foundation or frontier models, including LLMs
Combining Chain-of-Thought (CoT) with Reinforcement Learning (RL) improves text-to-image (T2I) generation, yet the underlying interaction between CoT's exploration and RL's optimization remains unclear. We present a systematic entropy-based analysis that yields three key insights: (1) CoT expands the generative explora...
[ "Language Models", "Autoregressive Image Generation", "Chain-of-Thought" ]
null
25,232
2604.02355
title_snapshot
[ 0.006094444543123245, -0.029477229341864586, -0.002713652327656746, 0.061201997101306915, 0.02817506529390812, 0.03720780089497566, 0.021453021094202995, -0.0042164865881204605, -0.02384175918996334, -0.04751933366060257, -0.046744901686906815, 0.018268024548888206, -0.06011936068534851, -...
Stopping Computation for Converged Tokens in Masked Diffusion-LM Decoding
https://openreview.net/forum?id=PzhNnMepgl
[ "Daisuke Oba", "Danushka Bollegala", "Masahiro Kaneko", "Naoaki Okazaki" ]
Poster
generative models
Masked Diffusion Language Models generate sequences via iterative sampling that progressively unmasks tokens. However, they still recompute the attention and feed-forward blocks for every token position at every step---even when many unmasked tokens are essentially fixed, resulting in substantial waste in compute. We p...
[ "diffusion language models", "compute efficient sampling", "skipping compute", "adaptive attention" ]
null
25,210
2602.06412
title_snapshot
[ -0.015658793970942497, -0.004160352982580662, -0.01170363463461399, 0.0468444786965847, 0.052689846605062485, 0.045342106372117996, 0.045058902353048325, 0.012855525128543377, -0.04397339001297951, -0.023670082911849022, 0.010647661983966827, -0.02963518351316452, -0.037528812885284424, -0...
Do LLM Agents Know How to Ground, Recover, and Assess? Evaluating Epistemic Competence in Information-Seeking Agents
https://openreview.net/forum?id=r0L9GwlnzP
[ "Jiaqi Shao", "Yuxiang Lin", "Munish Prasad Lohani", "Yufeng Miao", "Bing Luo" ]
Poster
datasets and benchmarks
Recent work has explored training Large Language Model (LLM) search agents with reinforcement learning (RL) for open-domain question answering. However, most evaluations focus solely on final answer accuracy, overlooking how these agents reason with and act on external evidence. We introduce **SeekBench**, the first pr...
[ "Epistemic Competence", "Evidence-Grounded Reasoning", "LLM Search Agents" ]
null
25,205
2509.22391
title_judge
[ -0.009243727661669254, -0.010033201426267624, -0.0011273273266851902, 0.05006920173764229, 0.06412498652935028, -0.00407753512263298, 0.03941287100315094, 0.007574622984975576, -0.01142845954746008, 0.0136251887306571, -0.04131830111145973, 0.05881056562066078, -0.04406445473432541, -0.025...
JailNewsBench: Multi-Lingual and Regional Benchmark for Fake News Generation under Jailbreak Attacks
https://openreview.net/forum?id=7dTqUaY2Kl
[ "Masahiro Kaneko", "Ayana Niwa", "Timothy Baldwin" ]
Poster
datasets and benchmarks
Fake news undermines societal trust and decision-making across politics, economics, health, and international relations, and in extreme cases threatens human lives and societal safety. Because fake news reflects region-specific political, social, and cultural contexts and is expressed in language, evaluating the risks ...
[ "fake news", "jailbreak", "llm", "multilingual" ]
null
25,200
2603.01291
title_snapshot
[ 0.007786114234477282, -0.03352055698633194, -0.030902119353413582, 0.04106951504945755, 0.03935965150594711, 0.024677636101841927, 0.0384380966424942, -0.011580457910895348, -0.035209771245718, -0.013985644094645977, -0.004565995652228594, 0.0013784753391519189, -0.09407519549131393, -0.01...
SRT: Super-Resolution for Time Series via Disentangled Rectified Flow
https://openreview.net/forum?id=I94Eg6cu7P
[ "Jufang Duan", "Shenglong Xiao", "Yuren Zhang" ]
Poster
learning on time series and dynamical systems
Fine-grained time series data with high temporal resolution is critical for accurate analytics across a wide range of applications. However, the acquisition of such data is often limited by cost and feasibility. This problem can be tackled by reconstructing high-resolution signals from low-resolution inputs based on sp...
[ "Time Series Super-Resolution", "Rectified Flow", "Temporal Disentanglement", "Implicit Neural Representations" ]
We propose SRT, a novel disentangled rectified flow framework for time series super-resolution that generates high-resolution details from low-resolution data, achieving state-of-the-art performance across nine benchmarks.
25,199
null
null
[ -0.008595719933509827, -0.011598201468586922, 0.0005305010126903653, 0.023685069754719734, 0.04803549125790596, 0.04382062703371048, 0.056847136467695236, -0.0010025723604485393, -0.024230027571320534, -0.03546677157282829, 0.0036181332543492317, -0.023784520104527473, -0.05144961178302765, ...
Conditioned Initialization for Attention
https://openreview.net/forum?id=cKNOCYPo2W
[ "Hemanth Saratchandran", "Simon Lucey" ]
Poster
other topics in machine learning (i.e., none of the above)
Transformers are a dominant architecture in modern machine learning, powering applications across vision, language, and beyond. At the core of their success lies the attention layer, where the query, key, and value matrices determine how token dependencies are captured. While considerable work has focused on scaling an...
[ "spectral conditioning transformers", "spectral properties of attention" ]
null
25,196
null
null
[ 0.00031346053583547473, -0.027473218739032745, 0.003993269056081772, 0.033552248030900955, 0.010580535978078842, 0.05270237848162651, 0.008758744224905968, -0.012775227427482605, -0.032771702855825424, -0.014339067041873932, -0.045101623982191086, 0.035214826464653015, -0.05935337767004967, ...
StreamingThinker: Large Language Models Can Think While Reading
https://openreview.net/forum?id=10Iiew095e
[ "Junlong Tong", "Yingqi Fan", "Anhao Zhao", "Yunpu Ma", "Xiaoyu Shen" ]
Poster
foundation or frontier models, including LLMs
Large language models (LLMs) have demonstrated remarkable capabilities in chain of thought (CoT) reasoning. However, the current LLM reasoning paradigm initiates thinking only after the entire input is available, which introduces unnecessary latency and weakens attention to earlier information in dynamic scenarios. Ins...
[ "LLMs", "Reasoning", "Streaming" ]
We propose StreamingThinker, a framework that enables LLMs to think while reading.
25,190
2510.17238
title_snapshot
[ -0.009013335220515728, -0.03380183130502701, -0.020035769790410995, 0.04450991749763489, 0.046600889414548874, 0.007504874840378761, 0.0010913159931078553, 0.02626020461320877, -0.04288908839225769, -0.0005076891975477338, -0.023575177416205406, 0.01664978265762329, -0.05484209582209587, -...
DeepCompress: A Dual Reward Strategy for Dynamically Exploring and Compressing Reasoning Chains
https://openreview.net/forum?id=K5A2jBmEBK
[ "Tian Liang", "Wenxiang Jiao", "Zhiwei He", "Jiahao Xu", "Haitao Mi", "Dong Yu" ]
Poster
reinforcement learning
Large Reasoning Models (LRMs) have demonstrated impressive capabilities but suffer from cognitive inefficiencies like ''overthinking'' simple problems and ''underthinking'' complex ones. While existing methods that use supervised fine-tuning (SFT) or reinforcement learning (RL) with token-length rewards can improve eff...
[ "Large Reasoning Models", "Reasoning Efficiency", "Reinforcement Learning" ]
This paper introduces DeepCompress, a dual reward strategy that simultaneously enhances both the accuracy and efficiency of large reasoning models.
25,170
2510.27419
title_snapshot
[ -0.022023998200893402, -0.03863639757037163, -0.02875700406730175, 0.04467054083943367, 0.06361613422632217, 0.042332738637924194, -0.007778821047395468, -0.004396781791001558, -0.032899949699640274, -0.0486149787902832, -0.025613421574234962, 0.033284008502960205, -0.028047440573573112, -...
Implicit 4D Gaussian Splatting for Fast Motion with Large Inter-Frame Displacements
https://openreview.net/forum?id=MWtXs60n38
[ "Seung-gyeom Kim", "Areum Kim", "Yongjae Yoo", "Sukmin Yun" ]
Poster
applications to computer vision, audio, language, and other modalities
Recent 4D Gaussian Splatting (4DGS) methods often fail under fast motion with large inter-frame displacements, where Gaussian attributes are poorly learned during training, and fast-moving objects are often lost from the reconstruction. In this work, we introduce Spatiotemporal Position Implicit Network for 4DGS, coine...
[ "4D Gaussian splatting", "4D reconstruction", "Dynamic rendering" ]
null
25,147
null
null
[ -0.020146874710917473, -0.01413054671138525, 0.029875602573156357, 0.06210993602871895, -0.0006022742600180209, 0.008527430705726147, 0.02965952642261982, 0.016459587961435318, -0.0223090797662735, -0.048326097428798676, 0.006083423737436533, -0.04653211310505867, -0.06050337105989456, 0.0...
HiDrop: Hierarchical Vision Token Reduction in MLLMs via Late Injection, Concave Pyramid Pruning, and Early Exit
https://openreview.net/forum?id=2baJBgfr9S
[ "Hao Wu", "Yingqi Fan", "Dai Jinyang", "Junlong Tong", "Yunpu Ma", "Xiaoyu Shen" ]
Poster
foundation or frontier models, including LLMs
The quadratic computational cost of processing vision tokens in Multimodal Large Language Models (MLLMs) hinders their widespread adoption. While progressive vision token pruning offers a promising solution, current methods misinterpret shallow layer functions and use rigid schedules, which fail to unlock the full effi...
[ "MLLMs", "Vision Token Pruning", "Efficiency and Compression", "Interpretability and Analysis" ]
null
25,145
2602.23699
title_snapshot
[ 0.006025304086506367, -0.00359352957457304, -0.001754006021656096, 0.04813256487250328, 0.004771657288074493, 0.03263311833143234, 0.033205337822437286, 0.003982188180088997, -0.038222651928663254, -0.02372201718389988, -0.013328558765351772, 0.009856005199253559, -0.061199579387903214, 0....
Why We Need New Benchmarks for Local Intrinsic Dimension Estimation
https://openreview.net/forum?id=ZEf03Uunvk
[ "Piotr Tempczyk", "Dominik Filipiak", "Łukasz Garncarek", "Ksawery Smoczyński", "Adam Kurpisz" ]
Poster
datasets and benchmarks
Neural Local Intrinsic Dimension (LID) estimators are typically bound to domain-specific architectures whose inductive biases can yield inconsistent estimates for the same underlying manifold. Existing evaluations either use overly simple synthetic data (with known LID) or real datasets (with unknown LID), obscuring tr...
[ "Local intrinsic dimension estimation", "LIDL", "FLIPD", "Diffusion Models", "Benhamark", "Normalizing Flows", "ESS", "Normal Bundle", "NB", "LID" ]
We show that LID estimation community needs new benchmarks for intrinsic dimension estimation and come to interesting conclusions on the performance of existing algorithms.
25,138
null
null
[ -0.033650197088718414, -0.002662369981408119, 0.007099318318068981, 0.05030860751867294, 0.011512598022818565, 0.050884418189525604, 0.047526150941848755, -0.025429269298911095, -0.022787094116210938, -0.039861809462308884, 0.01044313795864582, -0.015612120740115643, -0.08386032283306122, ...
MergePRAG: Orthogonal Merging of Passage-experts for Multi-hop Parametric RAG
https://openreview.net/forum?id=FSL1J2gmJV
[ "Xuebing Liu", "Shanbao Qiao", "Roseline Nyange", "Dongwook Min", "Hyun Kim", "Seung-Hoon Na" ]
Poster
applications to computer vision, audio, language, and other modalities
Large language models (LLMs) can be enhanced with external knowledge through two dominant approaches: (1) **retrieval-augmented generation (RAG)**, which supplements LLMs with in-context retrieved passages, and (2) **parametric knowledge adaptation (PKA)**, which directly updates model parameters with new domain knowle...
[ "Multi-hop reasoning", "Knowledge enhancement", "Retrieval-augmented generation", "Hypernetwork-based expert generation" ]
null
25,133
null
null
[ -0.0022862658370286226, -0.03219548985362053, 0.002636224962770939, 0.05402059108018875, 0.031538985669612885, 0.014526481740176678, 0.05483398213982582, -0.015635842457413673, -0.023322323337197304, -0.03228260949254036, -0.02017933689057827, 0.010296407155692577, -0.06398970633745193, -0...
Gradient-Based Diversity Optimization with Differentiable Top-$k$ Objective
https://openreview.net/forum?id=cuzWopwoZG
[ "Tianyi Zhou", "Sebastian Dalleiger", "Ece Calikus", "Aristides Gionis" ]
Poster
other topics in machine learning (i.e., none of the above)
Predicting relevance is a pervasive problem across digital platforms, covering social media, entertainment, and commerce. However, when optimized solely for relevance and engagement, many machine-learning models amplify data biases and produce homogeneous outputs, reinforcing filter bubbles and content uniformity. To a...
[ "Diversity Optimization", "Gradient-based learning", "Recommendation" ]
We introduce a differentiable top-k diversity objective with direct and indirect optimization, showing fine-tuning quickly adds diversity at scale with negligible accuracy loss.
25,123
null
null
[ -0.017101746052503586, -0.021601781249046326, 0.026115572080016136, 0.03574858605861664, 0.037916019558906555, 0.023651763796806335, 0.01979045942425728, -0.01523913349956274, -0.022881656885147095, -0.020351551473140717, -0.027021488174796104, 0.007343309000134468, -0.05874934419989586, -...
Understanding vs. Generation: Navigating Optimization Dilemma in Multimodal Models
https://openreview.net/forum?id=1smez00sCm
[ "Sen Ye", "Mengde Xu", "Shuyang Gu", "Di He", "Liwei Wang", "Winston Hu" ]
Poster
applications to computer vision, audio, language, and other modalities
Current research in multimodal models faces a key challenge where enhancing generative capabilities often comes at the expense of understanding, and vice versa. We analyzed this trade-off and identify the primary cause might be the potential conflict between generation and understanding, which creates a competitive dyn...
[ "Unified Multimodal Large Models", "Text-to-image generation", "Reasoning Models" ]
null
25,103
2602.15772
title_snapshot
[ -0.019158927723765373, -0.025874562561511993, 0.00817939080297947, 0.04621773585677147, 0.03565121442079544, 0.02193990908563137, 0.035847004503011703, 0.014621582813560963, -0.048410624265670776, -0.022963549941778183, -0.021911457180976868, 0.04136190935969353, -0.05863061174750328, -0.0...
When Language Models Lose Their Mind: The Consequences of Brain Misalignment
https://openreview.net/forum?id=MkrsbXl1GI
[ "Gabriele Merlin", "Mariya Toneva" ]
Poster
foundation or frontier models, including LLMs
While brain-aligned large language models (LLMs) have garnered attention for their potential as cognitive models and for potential for enhanced safety and trustworthiness in AI, the role of this brain alignment for linguistic competence remains uncertain. In this work, we investigate the functional implications of brai...
[ "language models", "brain alignment", "brain misalignment", "linguistic competence", "neuroscience", "fMRI" ]
null
25,096
2603.23091
title_snapshot
[ -0.021792680025100708, 0.01957767643034458, -0.020176967605948448, 0.004366629291325808, 0.022590678185224533, 0.022222764790058136, 0.04059366509318352, 0.05185503140091896, -0.048549946397542953, -0.0160202756524086, -0.03333670273423195, 0.003067456651479006, -0.06287327408790588, -0.00...
Resurfacing the Instance-only Dependent Label Noise Model through Loss Correction
https://openreview.net/forum?id=tuvkrivvbG
[ "Mustafa Enes Aydın", "Maarten De Vos", "Alexander Bertrand" ]
Poster
unsupervised, self-supervised, semi-supervised, and supervised representation learning
We investigate the label noise problem in supervised binary classification settings and resurface the underutilized instance-_only_ dependent noise model through loss correction. On the one hand, based on risk equivalence, the instance-aware loss correction scheme completes the bridge from _empirical noisy risk minimiz...
[ "label noise", "loss correction", "instance-dependence", "risk equivalence" ]
We resurrect the instance-only dependent label noise model via loss correction that connects the empirical-noisy-risk with the true-clean-risk.
25,088
null
null
[ -0.0030875185038894415, -0.0015195277519524097, -0.008353463374078274, 0.0403393879532814, 0.04370318725705147, 0.058977965265512466, 0.010470887646079063, -0.01698901876807213, -0.023372016847133636, -0.034520138055086136, -0.022494444623589516, 0.038486987352371216, -0.0683480054140091, ...
Animating the Uncaptured: Humanoid Mesh Animation with Video Diffusion Models
https://openreview.net/forum?id=DIPeQTxpe7
[ "Marc Benedí San Millán", "Angela Dai", "Matthias Nießner" ]
Poster
applications to computer vision, audio, language, and other modalities
Animation of humanoid characters is essential in various graphics applications, but require significant time and cost to create realistic animations. We propose an approach to synthesize 4D animated sequences of input static 3D humanoid meshes, leveraging strong generalized motion priors from generative video models --...
[ "Motion generation", "Motion Tracking & Transfer" ]
A method to animate humanoid meshes from a text prompt by transferring motion generated by video diffusion models to the mesh.
25,066
2503.15996
title_snapshot
[ 0.010473420843482018, 0.009663894772529602, 0.00011796601756941527, 0.05269072577357292, 0.03847963362932205, 0.03522751107811928, 0.007504709530621767, 0.009901599958539009, -0.042184483259916306, -0.049041952937841415, -0.03154423460364342, -0.026349740102887154, -0.05156901478767395, 0....
EMFuse: Energy-based Model Fusion for Decision Making
https://openreview.net/forum?id=6wDp8XRmNI
[ "Kejie He", "Yi-Chen Li", "Yang Yu" ]
Poster
reinforcement learning
Model fusion has emerged as a promising research direction, offering a resource-efficient paradigm that leverages existing pre-trained models to circumvent the need for training from scratch. In this work, we investigate the fusion of models specifically adapted for decision-making tasks. This challenge divides into tw...
[ "Model Fusion", "Energy-Based Model", "Decision Making" ]
null
25,065
null
null
[ -0.013526111841201782, -0.04152541980147362, -0.009391112253069878, 0.062373265624046326, 0.04411224648356438, 0.00350810750387609, 0.002236240776255727, 0.011148620396852493, -0.033890582621097565, -0.06087331101298332, 0.005090091843158007, -0.0032312485855072737, -0.09162340313196182, -...
The Unseen Bias: How Norm Discrepancy in Pre-Norm MLLMs Leads to Visual Information Loss
https://openreview.net/forum?id=GVVNG2EMQv
[ "Bozhou Li", "Xinda Xue", "Sihan Yang", "Yang Shi", "Xinlong Chen", "Yushuo Guan", "Yuanxing Zhang", "Wentao Zhang" ]
Poster
foundation or frontier models, including LLMs
Multimodal Large Language Models (MLLMs), which couple pre-trained vision encoders and language models, have shown remarkable capabilities. However, their reliance on the ubiquitous Pre-Norm architecture introduces a subtle yet critical flaw: a severe norm disparity between the high-norm visual tokens and the low-norm ...
[ "MultiModal Large Language Model;Pre-Normlization" ]
null
25,055
2512.08374
title_snapshot
[ 0.02783830463886261, -0.023589681833982468, 0.004909893032163382, 0.027720792219042778, 0.010330986231565475, -0.013483915477991104, 0.04247527942061424, 0.03212084248661995, -0.0559636652469635, -0.005846468731760979, -0.027747172862291336, 0.010815609246492386, -0.0696190744638443, -0.01...
ExPO-HM: Learning to Explain-then-Detect for Hateful Meme Detection
https://openreview.net/forum?id=bEejbORUI5
[ "Jingbiao Mei", "Mingsheng Sun", "Jinghong Chen", "Pengda Qin", "Yuhong Li", "Da Chen", "Bill Byrne" ]
Poster
applications to computer vision, audio, language, and other modalities
Hateful memes have emerged as a particularly challenging form of online abuse, motivating the development of automated detection systems. Most prior approaches rely on direct detection, producing only binary predictions. Such models fail to provide the context and explanations that real-world moderation requires. Recen...
[ "Hateful Meme Detection" ]
null
25,052
2510.08630
title_snapshot
[ 0.02292797900736332, -0.012223864905536175, -0.004507741425186396, 0.04033868759870529, 0.0027280873619019985, 0.00894194282591343, 0.020147835835814476, 0.002831747056916356, -0.030699491500854492, 0.0021036809775978327, -0.008638167753815651, 0.018078038468956947, -0.066596619784832, 0.0...
Recurrent Action Transformer with Memory
https://openreview.net/forum?id=kByN4v0M3e
[ "Egor Cherepanov", "Aleksei Staroverov", "Alexey Kovalev", "Aleksandr Panov" ]
Poster
reinforcement learning
Transformers have become increasingly popular in offline reinforcement learning (RL) due to their ability to treat agent trajectories as sequences, reframing policy learning as a sequence modeling task. However, in partially observable environments (POMDPs), effective decision-making depends on retaining information ab...
[ "RL", "Offline RL", "Memory", "Transformers", "POMDP" ]
The paper proposes Recurrent Action Transformer with Memory - a transformer model with recurrent memory and a procedure for training it for memory-intensive environments in an Offline RL setting.
25,049
2306.09459
title_snapshot
[ -0.03580843284726143, -0.03860992565751076, -0.020786378532648087, 0.03806352987885475, 0.02919887937605381, 0.021178418770432472, 0.013589871115982533, 0.019797900691628456, -0.04817744717001915, 0.0006016987608745694, -0.0240156352519989, 0.03149154409766197, -0.06621947884559631, -0.013...
Exploring Diverse Generation Paths via Inference-time Stiefel Activation Steering
https://openreview.net/forum?id=v0QOVSVPtq
[ "Dongxuan Zhu", "Ly Tran Ho Khanh", "Andy Yat-Ming Cheung", "Man-Chung Yue", "Viet Anh Nguyen" ]
Poster
optimization
Language models often default to a narrow set of high-probability outputs, leaving their generation paths homogeneous and prone to mode collapse. Sampling-based strategies inject randomness but still struggle to guarantee diversity across multiple concurrent generation runs. We address this limitation by introducing ST...
[ "activation steering", "generation diversity", "manifold opimization" ]
null
25,047
2601.22010
title_snapshot
[ -0.02973194792866707, -0.0072205886244773865, 0.013035044074058533, 0.026445923373103142, 0.013632270507514477, 0.02959481067955494, 0.053214892745018005, 0.008435626514256, -0.006807101424783468, 0.0013208569725975394, -0.02873809076845646, 0.015147191472351551, -0.06197783350944519, 0.01...
Random Controlled Differential Equations
https://openreview.net/forum?id=kHqt0ZSbKT
[ "Francesco Piatti", "Thomas Cass", "William F. Turner" ]
Poster
learning on time series and dynamical systems
We introduce a training-efficient framework for time-series learning in which large randomly parameterized controlled and rough differential equations act as continuous-time reservoirs. These random dynamical systems map input paths to rich path-dependent representations, while only a linear readout layer is trained, y...
[ "random features", "time-series", "path signatures", "CDEs", "RDEs", "reservoir computing", "kernels" ]
null
25,045
2512.23670
title_snapshot
[ -0.03655446320772171, -0.021137477830052376, 0.010639887303113937, 0.0762919932603836, 0.024664675816893578, 0.03301927074790001, 0.03039541095495224, 0.00594785762950778, -0.005075701046735048, -0.04522068053483963, 0.02725035883486271, 0.0015631727874279022, -0.05746834725141525, -0.0065...
FastFlow: Accelerating The Generative Flow Matching Models with Bandit Inference
https://openreview.net/forum?id=wWkyL8D9xd
[ "Divya Jyoti Bajpai", "Dhruv Bhardwaj", "Soumya Roy", "Tejas Duseja", "Harsh Agarwal", "Aashay Sandansing", "Manjesh Kumar Hanawal" ]
Poster
generative models
Flow-matching models deliver state-of-the-art fidelity in image and video generation, but the inherent sequential denoising process renders them slower. Existing acceleration methods like distillation, trajectory truncation, and consistency approaches are static, require retraining, and often fail to generalize across ...
[ "generative modelling", "faster inference." ]
Adaptive inference method for accelerating flow matching based visual generation.
25,044
2602.11105
title_snapshot
[ 0.019521670415997505, -0.03303475305438042, -0.014679498970508575, 0.07919667661190033, 0.021135549992322922, 0.05157000944018364, 0.022136949002742767, 0.013075384311378002, -0.0027911169454455376, -0.05616479739546776, -0.016756799072027206, -0.021603737026453018, -0.07457233220338821, -...
Memory, Benchmark & Robots: A Benchmark for Solving Complex Tasks with Reinforcement Learning
https://openreview.net/forum?id=9cLPurIZMj
[ "Egor Cherepanov", "Nikita Kachaev", "Alexey Kovalev", "Aleksandr Panov" ]
Poster
applications to robotics, autonomy, planning
Memory is crucial for enabling agents to tackle complex tasks with temporal and spatial dependencies. While many reinforcement learning (RL) algorithms incorporate memory, the field lacks a universal benchmark to assess an agent's memory capabilities across diverse scenarios. This gap is particularly evident in tableto...
[ "Memory", "Benchmark", "Robots", "POMDP", "RL" ]
A benchmark of 32 memory tasks for tabletop robotic manipulation, a benchmark to test the memory of an RL agent and classification of memory tasks in RL by type of memory usage
25,030
2502.10550
title_snapshot
[ -0.04040532186627388, -0.0011343987425789237, -0.011535354889929295, 0.04129457473754883, 0.01930316351354122, 0.014617197215557098, 0.029400503262877464, 0.009375744499266148, -0.05514215677976608, -0.02946387603878975, -0.010507697239518166, 0.017854951322078705, -0.06212197244167328, -0...
Unraveling the Complexity of Memory in RL Agents: an Approach for Classification and Evaluation
https://openreview.net/forum?id=lJKdOYFF5W
[ "Egor Cherepanov", "Nikita Kachaev", "Artem Zholus", "Alexey Kovalev", "Aleksandr Panov" ]
Poster
reinforcement learning
The incorporation of memory into agents is essential for numerous tasks within the domain of Reinforcement Learning (RL). In particular, memory is paramount for tasks that require the use of past information, adaptation to novel environments, and improved sample efficiency. However, the term ``memory'' encompasses a wi...
[ "RL", "POMDP", "Memory", "Classification" ]
A formal description of the memory types of RL agents and a methodology for conducting an experiment to test the memory.
25,014
2412.06531
title_snapshot
[ -0.0495535284280777, -0.022590192034840584, -0.019476288929581642, 0.03207409381866455, 0.04556538909673691, 0.0024412120692431927, 0.023231418803334236, 0.004628431051969528, -0.05064670369029045, -0.025950409471988678, -0.017118461430072784, 0.036203350871801376, -0.06919434666633606, -0...
ELMUR: External Layer Memory with Update/Rewrite for Long-Horizon RL Problems
https://openreview.net/forum?id=bm3rbtEMFj
[ "Egor Cherepanov", "Alexey Kovalev", "Aleksandr Panov" ]
Poster
reinforcement learning
Real-world robotic agents must act under partial observability and long horizons, where key cues may appear long before they affect decision making. However, most modern approaches rely solely on instantaneous information, without incorporating insights from the past. Standard recurrent or transformer models struggle w...
[ "RL", "POMDP", "Memory", "Transformer", "Robotics" ]
ELMUR is a transformer model with layer-local external memory and LRU-based memory updates for long-horizon reasoning in POMDPs
25,001
2510.07151
title_snapshot
[ -0.05156475305557251, -0.030418576672673225, 0.0036735835019499063, 0.039100468158721924, 0.01899983175098896, 0.012130515649914742, -0.0050865658558905125, 0.028417956084012985, -0.031992215663194656, -0.029794661328196526, -0.04656391963362694, 0.017857925966382027, -0.08437781780958176, ...
Improving Extreme Wind Prediction with Frequency-Informed Learning
https://openreview.net/forum?id=IJAPVmxQYU
[ "Chenrui Xu", "Xi Huang", "Ying-Jun Angela Zhang", "Jianwei Huang" ]
Poster
applications to physical sciences (physics, chemistry, biology, etc.)
Accurate prediction of extreme wind velocities has substantial significance in industry, particularly for the operation management of wind power plants. Although the state-of-the-art data-driven models perform well for general meteorological forecasting, they may exhibit large errors for extreme weather—for example, sy...
[ "Extreme Weather Forecasting", "Meteorological Analysis", "AI for Science" ]
null
25,000
null
null
[ -0.011229163967072964, -0.041117724031209946, 0.013888837769627571, 0.03645430877804756, 0.025351114571094513, -0.013889108784496784, 0.01032908447086811, -0.007317675277590752, -0.043370846658945084, -0.02390548586845398, -0.02035994455218315, 0.011069054715335369, -0.03750895708799362, 0...
Reinforcement Learning Fine-Tuning Enhances Activation Intensity and Diversity in the Internal Circuitry of LLMs
https://openreview.net/forum?id=tzS9roOTdj
[ "Honglin Zhang", "Qianyue Hao", "Fengli Xu", "Yong Li" ]
Poster
interpretability and explainable AI
Large language models (LLMs) acquire extensive prior knowledge through large-scale pretraining and can be further enhanced via supervised fine-tuning (SFT) or reinforcement learning (RL)-based post-training. A growing body of evidence has shown that RL fine-tuning improves the capability of LLMs beyond what SFT alone a...
[ "Large Language Models; Reinforcement Learning Fine-Tuning; Edge Attribution Patching" ]
This work utilizes edge attribution patching (EAP) to investigate the internal differences of LLMs before and after RL fine-tuning, and uncovers that RL enhances activation intensity and diversity in the internal circuitry of LLMs.
24,998
2509.21044
title_snapshot
[ -0.017319973558187485, -0.022050537168979645, 0.0035997238010168076, 0.018706856295466423, 0.05813571438193321, 0.015353124588727951, 0.035056889057159424, 0.02051353268325329, -0.052084535360336304, -0.013931788504123688, -0.02048628218472004, 0.05147451162338257, -0.05776321142911911, -0...
Benchmarking Stochastic Approximation Algorithms for Fairness-Constrained Training of Deep Neural Networks
https://openreview.net/forum?id=JxmjzC6syB
[ "Andrii Kliachkin", "Jana Lepšová", "Gilles Bareilles", "Jakub Marecek" ]
Poster
datasets and benchmarks
The ability to train Deep Neural Networks (DNNs) with constraints is instrumental in improving the fairness of modern machine-learning models. Many algorithms have been analysed in recent years, and yet there is no standard, widely accepted method for the constrained training of DNNs. In this paper, we provide a challe...
[ "Fair Machine Learning", "stochastic approximation", "Augmented Lagrangian", "Sequential Quadratic Programming", "benchmarking" ]
We provide a benchmark for comparing stochastic approximation algorithms, based on real-world fairness-constrained learning problems.
24,989
2507.04033
title_snapshot
[ -0.025741610676050186, -0.020827200263738632, -0.04996085911989212, 0.04427905008196831, 0.01870601624250412, 0.04161740466952324, 0.024663245305418968, 0.002926850225776434, -0.04602229595184326, -0.025572262704372406, 0.004618118517100811, 0.002889573108404875, -0.07237689197063446, -0.0...
Automated Stateful Specialization for Adaptive Agent Systems
https://openreview.net/forum?id=UESTP6dR1K
[ "Myan Vu", "Harrish Ayyanar", "PANG JIANG", "Anwiketh Reddy", "Mayank Goel" ]
Poster
foundation or frontier models, including LLMs
Current automated agent design frameworks produce either static workflows that lack adaptability or per-query optimizers that prevent the accumulation of deep, agent-level task expertise. We propose a new direction that reconciles these paradigms: creating stateful teams of specialist agents that accumulate knowledge o...
[ "LLMs", "Autonomous Agents", "Agent Specialization" ]
We introduce a framework that creates persistent, specialist agent teams through an offline lifecycle of discovery and cultivation, and deploys them with an online policy that efficiently adapts the team's structure for novel tasks.
24,986
null
null
[ -0.008958926424384117, -0.030656062066555023, -0.002983870217576623, 0.034232139587402344, 0.05962313339114189, 0.01367120910435915, 0.035207491368055344, -0.011883866041898727, -0.012028218246996403, -0.04910307005047798, -0.03745955973863602, 0.02631770260632038, -0.08879180997610092, -0...
Adaptive Methods Are Preferable in High Privacy Settings: An SDE Perspective
https://openreview.net/forum?id=hSpA4DAoMk
[ "Enea Monzio Compagnoni", "Alessandro Stanghellini", "Rustem Islamov", "Aurelien Lucchi", "Anastasia Koloskova" ]
Poster
optimization
Differential Privacy (DP) is becoming central to large-scale training as privacy regulations tighten. We revisit how DP noise interacts with _adaptivity_ in optimization through the lens of _stochastic differential equations_, providing the first SDE-based analysis of private optimizers. Focusing on *DP-SGD* and *DP-Si...
[ "Stochastic Differential Equations", "Differential Privacy" ]
With SDEs, we show that while DP-SignSGD is better under tight privacy or noisy batches, DP-SGD is better otherwise, and adaptivity needs far less hyperparameter tuning across privacy levels.
24,978
2603.03226
title_snapshot
[ -0.025433195754885674, 0.010860895738005638, 0.011530370451509953, 0.06608046591281891, 0.04661788046360016, 0.044794704765081406, 0.05643213912844658, -0.018937965855002403, 0.0056143710389733315, -0.051363371312618256, 0.00885287020355463, -0.00536278635263443, -0.054181091487407684, 0.0...
From Parameters to Behaviors: Unsupervised Compression of the Policy Space
https://openreview.net/forum?id=VqnBaeu43F
[ "Davide Tenedini", "Riccardo Zamboni", "Mirco Mutti", "Marcello Restelli" ]
Poster
reinforcement learning
Despite its recent successes, Deep Reinforcement Learning (DRL) is notoriously sample-inefficient. We argue that this inefficiency stems from the standard practice of optimizing policies directly in the high-dimensional and highly redundant parameter space $\\Theta$. This challenge is greatly compounded in multi-task s...
[ "reinforcement learning", "unsupervised reinforcement learning", "unsupervised representation learning" ]
null
24,961
2509.22566
title_snapshot
[ -0.023144753649830818, -0.041007671505212784, -0.028235556557774544, 0.043306440114974976, 0.040226951241493225, 0.0590645894408226, 0.01438215933740139, -0.032299600541591644, -0.0055398219265043736, -0.04961404576897621, -0.006213975138962269, -0.003951634280383587, -0.07449852675199509, ...
Best-of-Infinity: Asymptotic Performance of Test-Time LLM Ensembling
https://openreview.net/forum?id=3qiCnLf3jf
[ "Junpei Komiyama", "Daisuke Oba", "Masafumi Oyamada" ]
Poster
foundation or frontier models, including LLMs
We study best-of-$N$ for large language models (LLMs) where the selection is based on majority voting. In particular, we analyze the limit $N \to \infty$, which we denote as best-of-$\infty$. While this approach achieves impressive performance in the limit, it requires an infinite test-time budget. To address this, we ...
[ "LLM", "test-time compute", "majority voting", "LLM ensemble" ]
null
24,953
2509.21091
title_judge
[ -0.036690905690193176, -0.03522758185863495, -0.011418774724006653, 0.023030759766697884, 0.04479706659913063, 0.014869490638375282, 0.028842896223068237, 0.012107664719223976, -0.01263924315571785, 0.003994620870798826, -0.018411125987768173, 0.011504960246384144, -0.07133583724498749, -0...
Copy-Paste to Mitigate Large Language Model Hallucinations
https://openreview.net/forum?id=crKJJ4Ej60
[ "Yongchao Long", "Yingying Zhang", "Xianbin Wen", "Xian Wu", "Yuxi Zhou", "Shenda Hong" ]
Poster
foundation or frontier models, including LLMs
While Retrieval-Augmented Generation (RAG) enables large language models (LLMs) to generate contextually grounded responses, contextual faithfulness remains challenging as LLMs may not consistently trust provided context, leading to hallucinations that undermine reliability. We observe an inverse correlation between re...
[ "Hallucination", "Context Learning", "Contextual Faithfulness", "Knowledge Conflict", "Model Interpretability" ]
We propose Copy-Paste, a paradigm embedding contextual fragments for faithfulness, instantiated as CopyPasteLLM—achieving 12.2-24.5% accuracy gains with only 365 samples (1/50th of baseline) by recalibrating parametric knowledge.
24,938
2510.00508
title_snapshot
[ -0.0007976432098075747, 0.012959836982190609, -0.004038218874484301, 0.047245875000953674, 0.01162046194076538, 0.015174798667430878, 0.04727179557085037, 0.018194595351815224, -0.036781273782253265, -0.034770287573337555, -0.06520821899175644, 0.02914968505501747, -0.07877283543348312, -0...
Programming by Backprop: An Instruction is Worth 100 Examples When Finetuning LLMs
https://openreview.net/forum?id=y1OWj26FCo
[ "Jonathan Cook", "Silvia Sapora", "Arash Ahmadian", "Akbir Khan", "Tim Rocktäschel", "Jakob Nicolaus Foerster", "Laura Ruis" ]
Poster
foundation or frontier models, including LLMs
Large language models (LLMs) are typically trained to acquire behaviours from demonstrations or experience, yet much of their training data is declarative: instructions, rules, and descriptions that specify behaviours without showing how to execute them. We introduce **Programming by Backprop (PBB)**: a training regime...
[ "Large Language Models", "Abstraction", "Procedural Knowledge" ]
LLMs can learn to execute procedures that are described symbolically in their training data, but only with specific finetuning curricula.
24,916
2506.18777
title_snapshot
[ -0.03932791203260422, -0.02944139763712883, -0.0377829410135746, 0.012000530026853085, 0.050256017595529556, 0.012887630611658096, 0.043430134654045105, 0.011793929152190685, -0.02336747758090496, -0.0023775824811309576, -0.03871874511241913, 0.050520479679107666, -0.06841406226158142, -0....
Aria: an Agent for Retrieval and Iterative Auto-Formalization via Dependency Graph
https://openreview.net/forum?id=CPxZClPMiy
[ "Hanyu Wang", "Ruohan Xie", "Yutong Wang", "Guoxiong Gao", "XintaoYu", "Bin Dong" ]
Poster
foundation or frontier models, including LLMs
Accurate auto-formalization of theorem statements is essential for advancing automated discovery and verification of research-level mathematics, yet remains a major bottleneck for LLMs due to hallucinations, semantic mismatches, and their inability to synthesize new definitions. To tackle these issues, we present Aria ...
[ "Lean 4", "Autoformalization", "LLM", "Graph-of-Thought", "Retrieval Augmented Generation" ]
null
24,910
2510.04520
title_snapshot
[ -0.02139328606426716, -0.012103226035833359, -0.005574331618845463, 0.03644527122378349, 0.03340645134449005, 0.01015305332839489, 0.017655756324529648, -0.0021517921704798937, -0.004169783089309931, 0.0052204676903784275, -0.014760852791368961, 0.029111705720424652, -0.051236458122730255, ...
Complexity- and Statistics-Guided Anomaly Detection in Time Series Foundation Models
https://openreview.net/forum?id=rBt9aW3Mx7
[ "Jongwon Kim", "Samuel Yoon", "Young Myoung Ko", "Yerin Kim", "Sung Il Kim", "JAEUNG TAE" ]
Poster
learning on time series and dynamical systems
This paper introduces a methodology for anomaly detection in time series using Time Series Foundation Models (TFMs). While TFMs have achieved strong success in forecasting, their role in anomaly detection remains underexplored. We identify two key challenges when applying TFMs to reconstruction-based anomaly detection ...
[ "Timeseries anomaly detection", "Timeseries foundation model", "Reconstruction based anomaly detection" ]
We propose solutions based on a complexity measure α that captures high-frequency complexity and restores statistical features removed by RevIN, leading to theoretical and empirical improvements in anomaly detection.
24,889
null
null
[ -0.012637798674404621, -0.05377160385251045, 0.0039459411054849625, 0.006284254603087902, 0.07680746167898178, 0.03603588044643402, 0.034922461956739426, -0.015688125044107437, -0.048621587455272675, -0.048745304346084595, -0.007649508770555258, 0.009472510777413845, -0.054752714931964874, ...
Learning Posterior Predictive Distributions for Node Classification from Synthetic Graph Priors
https://openreview.net/forum?id=FmxRzlu0rT
[ "Jeongwhan Choi", "Jongwoo Kim", "Woosung Kang", "Noseong Park" ]
Poster
learning on graphs and other geometries & topologies
One of the most challenging problems in graph machine learning is generalizing across graphs with diverse properties. Graph neural networks (GNNs) face a fundamental limitation: they require separate training for each new graph, preventing universal generalization across diverse graph datasets. A critical challenge fac...
[ "graph machine learning", "node classification" ]
null
24,880
2604.19028
title_snapshot
[ -0.009741176851093769, -0.016362549737095833, 0.009760233573615551, 0.04745844379067421, 0.03637662157416344, 0.033709242939949036, 0.014696537517011166, 0.0027536393608897924, -0.014948359690606594, -0.02688175067305565, 0.0031547038815915585, -0.0063258688896894455, -0.07262738794088364, ...
Dynamical properties of dense associative memory
https://openreview.net/forum?id=TeDkzf34hs
[ "Kazushi Mimura", "Junichi Takeuchi", "Yuto Sumikawa", "Yoshiyuki Kabashima", "Anthony CC Coolen" ]
Poster
learning theory
Dense associative memory, a fundamental instance of modern Hopfield networks, can store a large number of memory patterns as equilibrium states of recurrent networks. While the stationary-state storage capacity has been investigated, its dynamical properties have not yet been discussed. In this paper, we analyze the dy...
[ "Hopfield networks", "dense associative memory", "dynamics", "convergence time", "attraction basin", "generating functional analysis" ]
We analyze the recalling process using the generating functional analysis and discuss the convergence time and the attraction basins and so on.
24,879
2506.00851
title_snapshot
[ -0.049683887511491776, 0.0019733814988285303, 0.006268537603318691, 0.04124201834201813, 0.029798319563269615, 0.005531554576009512, 0.013043428771197796, 0.02346229925751686, -0.05700857937335968, -0.02135852724313736, 0.024314217269420624, -0.020991407334804535, -0.062067486345767975, 0....
Detection of unknown unknowns in autonomous systems
https://openreview.net/forum?id=GrsofC2FqF
[ "Ayan Banerjee", "Sandeep Gupta" ]
Poster
probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)
Unknown unknowns (U2s) are deployment-time scenarios absent from development/testing. Unlike conventional anomalies, U2s are not out-of-distribution (OOD); they stem from changes in underlying system dynamics without a distribution shift from normal data. Thus, existing multi-variate time series anomaly detection (MTAD...
[ "unknown unknowns", "autonomous systems", "conformal bounds" ]
We formalize U2 (non-OOD dynamic changes without distribution shift), release 8 U2 benchmarks, and propose SPIE-AD—a zero-shot U2 detection method.
24,864
null
null
[ -0.006651714909821749, 0.001191331073641777, -0.026995375752449036, 0.06062847003340721, 0.07231513410806656, 0.018991032615303993, 0.03131409361958504, -0.010655169375240803, -0.04915010929107666, -0.04482800140976906, -0.02063293382525444, 0.028006860986351967, -0.05305854603648186, -0.0...
COSMO-INR: Complex Sinusoidal Modulation for Implicit Neural Representations
https://openreview.net/forum?id=yGJrvSU6wK
[ "Pandula Thennakoon", "Avishka Ranasinghe", "Mario De Silva", "Buwaneka Epakanda", "Roshan Godaliyadda", "Mervyn Parakrama Bandara Ekanayake", "Vijitha R. Herath" ]
Poster
unsupervised, self-supervised, semi-supervised, and supervised representation learning
Implicit neural representations (INRs) have recently emerged as a powerful paradigm for modeling data, offering a continuous alternative to traditional discrete signal representations. Their ability to compactly encode complex signals has led to strong performance across a wide range of computer vision tasks. In previo...
[ "Implicit Neural Networks", "Chebyshev Polynomials", "Raised cosine filter", "Spectral bias" ]
A study on the effect of complex sinusoidal modulation on INR activation functions
24,849
2505.11640
title_snapshot
[ -0.0011613521492108703, -0.014233901165425777, 0.0031247097067534924, 0.0037649802397936583, 0.02104146219789982, 0.029708437621593475, -0.0008749720291234553, 0.007116574794054031, -0.040942877531051636, -0.02752644568681717, -0.011470145545899868, 0.0008379827486351132, -0.0609690360724926...
Enough is as good as a feast: A Comprehensive Analysis of How Reinforcement Learning Mitigates Task Conflicts in LLMs
https://openreview.net/forum?id=N4l4Jp50R4
[ "Zixuan Ren", "Jinliang Lu", "Junhong Wu", "Yang Zhao", "Dai Dai", "Hua Wu", "Haifeng Wang", "Chengqing Zong" ]
Poster
foundation or frontier models, including LLMs
Model merging plays a crucial role in consolidating multiple specialized models into a single, unified model, especially in the era of large language models (LLMs). Recent research has primarily focused on developing strategies to enhance merging performance with the trained models, while the impact of training paradig...
[ "Large language model", "Reinforcement learning", "Model merging" ]
We provide a comprehensive analysis of how reinforcement learning mitigates task conflicts in LLMs
24,847
null
null
[ -0.04189014062285423, -0.0452047660946846, -0.008197644725441933, 0.023888615891337395, 0.05166010931134224, -0.010315298102796078, 0.03963945806026459, 0.012380613014101982, -0.043349489569664, -0.04847900569438934, -0.025771351531147957, 0.05115514248609543, -0.08588550239801407, -0.0221...
TiTok: Transfer Token-level Knowledge via Contrastive Excess to Transplant LoRA
https://openreview.net/forum?id=0B5K9pIdSK
[ "ChanJoo Jung", "Jaehyung Kim" ]
Poster
foundation or frontier models, including LLMs
Large Language Models (LLMs) are widely applied in real world scenarios, yet fine-tuning them comes with significant computational and storage costs. Parameter-Efficient Fine-Tuning (PEFT) methods such as LoRA mitigate these costs; however, the adapted parameters are dependent on the base model and cannot be transferre...
[ "Large Language Models", "Knowledge Transfer", "PEFT" ]
We propose a new framework TiTok, which enables effective LoRA transplantation through token-level knowledge transfer
24,843
2510.04682
title_snapshot
[ 0.011040209792554379, -0.05005701258778572, -0.020639246329665184, 0.05194889381527901, 0.032297249883413315, 0.03414132446050644, 0.02218911238014698, 0.027741245925426483, -0.028418967500329018, 0.010022321715950966, -0.0020749985706061125, 0.01423551794141531, -0.048334185034036636, -0....
Scalable Multilingual Multimodal Machine Translation with Speech-Text Fusion
https://openreview.net/forum?id=HQMVRQUEaM
[ "Yexing Du", "Youcheng Pan", "Zekun Wang", "Zheng Chu", "Yichong Huang", "Kaiyuan Liu", "Bo Yang", "Yang Xiang", "Ming Liu", "Bing Qin" ]
Poster
applications to computer vision, audio, language, and other modalities
Multimodal Large Language Models (MLLMs) have achieved notable success in enhancing translation performance by integrating multimodal information. However, existing research primarily focuses on image-guided methods, whose applicability is constrained by the scarcity of multilingual image-text pairs. The speech modalit...
[ "Speech", "Multimodal Machine Translation" ]
A Speech-guided framework that leverages synthetic speech and a self-evolution mechanism to improve translation
24,834
2602.21646
title_snapshot
[ 0.003262194339185953, -0.04199308529496193, 0.01305773202329874, 0.0339415967464447, 0.03556763008236885, 0.023683132603764534, 0.041437260806560516, 0.0310309287160635, -0.0270910132676363, -0.0026866456028074026, -0.016620805487036705, 0.029424350708723068, -0.06364905834197998, -0.00030...
From movement to cognitive maps: recurrent neural networks reveal how locomotor development shapes hippocampal spatial coding
https://openreview.net/forum?id=8bM7MkxJee
[ "Marco P Abrate", "Laurenz Muessig", "Joshua P Bassett", "Hui Min Tan", "Francesca Cacucci", "Thomas Joseph Wills", "Caswell Barry" ]
Oral
applications to neuroscience & cognitive science
The hippocampus contains neurons whose firing correlates with an animal's location and orientation in space. Collectively, these neurons are held to support a cognitive map of the environment, enabling the recall of and navigation to specific locations. Although recent studies have characterised the timelines of spatia...
[ "recurrent neural network", "spatial representations", "hippocampus", "development", "locomotion", "rats" ]
null
24,802
null
null
[ 0.0009668374550528824, 0.01309154275804758, -0.03144184872508049, 0.0057168323546648026, 0.05762358382344246, 0.027736889198422432, 0.03291922062635422, 0.0238534864038229, -0.04954875633120537, -0.06952716410160065, -0.030434822663664818, -0.03550633415579796, -0.036257218569517136, 0.007...
Computer Agent Arena: Toward Human-Centric Evaluation and Analysis of Computer-Use Agents
https://openreview.net/forum?id=3x4SDbXbgl
[ "Bowen Wang", "Xinyuan Wang", "Jiaqi Deng", "Tianbao Xie", "Ryan Li", "Yanzhe Zhang", "Junli Wang", "Dunjie Lu", "Zicheng Gong", "Gavin Li", "Toh Jing Hua", "Wei-Lin Chiang", "Ion Stoica", "Diyi Yang", "Yu Su", "Yi Zhang", "Zhiguo Wang", "Victor Zhong", "Tao Yu" ]
Poster
datasets and benchmarks
As Computer-Use Agents (CUAs) proliferate and grow increasingly capable, evaluation has become more challenging: static, manually curated benchmarks are narrow in domain, contamination-prone, and environment-heavy, and they diverge substantially from user-driven, real-world evaluation. We present Computer Agent Arena, ...
[ "Computer-Use Agent", "Visual Language Model", "Human-in-the-loop", "Evaluation" ]
null
24,792
null
null
[ -0.001403665286488831, -0.003641917835921049, -0.019489379599690437, 0.02397751808166504, 0.037395648658275604, -0.002306070178747177, 0.022318406030535698, 0.046984750777482986, -0.0061309742741286755, -0.043661728501319885, -0.012015377171337605, 0.03657098859548569, -0.0665242001414299, ...
Information-Theoretic Membership Inference for Granular Quantification of Memorization
https://openreview.net/forum?id=4KVeb0Vv13
[ "Jiashu Tao", "Reza Shokri" ]
Poster
alignment, fairness, safety, privacy, and societal considerations
Machine learning models are known to leak sensitive information, as they inevitably memorize (parts of) their training data. This risk is amplified for large language models (LLMs), which are trained on massive corpora and therefore create a more urgent need for privacy assessment prior to release. The standard method ...
[ "membership inference attack", "mia", "privacy", "llm privacy", "memorization" ]
We introduce a new state-of-the-art membership inference attack, InfoRMIA that dominates RMIA on all benchmarks. We also propose a token-level attack framework that has high power and can pinopint info leakage to token levels.
24,791
null
null
[ -0.014003388583660126, -0.009611609391868114, -0.02213412895798683, 0.03885495290160179, 0.03681959956884384, 0.01237280573695898, 0.051177557557821274, -0.008424497209489346, -0.04714907705783844, 0.032746538519859314, 0.001701792934909463, 0.02358463779091835, -0.04884086549282074, -0.00...
LS-Merge: Merging Language Models in Latent Space
https://openreview.net/forum?id=VSDV0SWwOC
[ "Bedionita Soro", "Aoxuan Silvia Zhang", "Bruno Andreis", "Jaehyeong Jo", "Song Chong", "Sung Ju Hwang" ]
Poster
generative models
Model merging in weight space is an efficient way to reuse pretrained models, but existing methods typically assume matching architectures or sizes, making heterogeneous merges brittle or infeasible. We address this limitation by encoding model weights into a smooth latent space, enabling cross-architecture operations,...
[ "LS-Merge", "LLM merging", "latent space", "weight space learning" ]
Merging Language Models in Latent Space
24,777
null
null
[ -0.01879391446709633, -0.03216990828514099, -0.03396449238061905, 0.050735730677843094, 0.04361089691519737, 0.06068941950798035, 0.01250507216900587, 0.006759960204362869, -0.021878991276025772, -0.038567762821912766, -0.028748370707035065, -0.015239902772009373, -0.07654563337564468, -0....
RLVER: Reinforcement Learning with Verifiable Emotion Rewards for Empathetic Agents
https://openreview.net/forum?id=P7wBg0vPTh
[ "Peisong Wang", "Ruotian Ma", "Bang Zhang", "Xingyu Chen", "Zhiwei He", "Kang Luo", "Qingsong Lv", "Qingxuan Jiang", "Zheng Xie", "Shanyi Wang", "CIXING LI", "Yuan Li", "Fanghua Ye", "Jian Li", "Yifan Yang", "Jia Li", "Zhaopeng Tu", "Xiaolong Li" ]
Poster
reinforcement learning
Large language models (LLMs) excel at logical and algorithmic reasoning, yet their emotional intelligence (EQ) still lags far behind their cognitive prowess. While reinforcement learning from verifiable rewards (RLVR) has advanced in other domains, its application to dialogue—especially for emotional intelligence—rema...
[ "Large language models", "Reinforcement Learning", "Agent" ]
null
24,774
2507.03112
title_snapshot
[ -0.005941914394497871, -0.028616219758987427, -0.00026936089852824807, 0.04303141310811043, 0.019785650074481964, 0.042818401008844376, 0.00945981964468956, 0.0068810428492724895, -0.01611511968076229, -0.028392642736434937, -0.04238717257976532, 0.04549558460712433, -0.04933592677116394, ...
Minor First, Major Last: A Depth-Induced Implicit Bias of Sharpness-Aware Minimization
https://openreview.net/forum?id=ErnnE2UNI2
[ "Chaewon Moon", "Dongkuk Si", "Chulhee Yun" ]
Poster
optimization
We study the implicit bias of sharpness-aware minimization (SAM) when training $L$-layer linear diagonal networks on linearly separable binary classification. For linear models ($L=1$), both $\ell_\infty$- and $\ell_2$-SAM recover the $\ell_2$ max-margin classifier, matching gradient descent (GD). However, for depth $L...
[ "sharpness-aware minimization", "implicit bias", "gradient flow" ]
null
24,773
2603.08290
title_snapshot
[ -0.033829353749752045, -0.009372660890221596, 0.01574953831732273, 0.029102636501193047, 0.007631830405443907, 0.031495869159698486, 0.03613591566681862, 0.01969618909060955, -0.028447574004530907, -0.0410919152200222, -0.015555710531771183, 0.011064549908041954, -0.058860715478658676, 0.0...
Is Graph Unlearning Ready for Practice? A Benchmark on Efficiency, Utility, and Forgetting
https://openreview.net/forum?id=gSPkuTTWgU
[ "Samyak Jain", "Ronak Kalvani", "sainyam galhotra", "Sayan Ranu" ]
Poster
datasets and benchmarks
Graph Neural Networks (\textsc{Gnn}s) are increasingly being deployed in sensitive, user-centric applications where regulations such as the GDPR mandate the ability to remove data upon request. This has spurred interest in graph unlearning, the task of removing the influence of specific training data from a trained \te...
[ "graph unlearning", "GNN", "graph neural network" ]
Our benchmark shows that graph unlearning can rival retraining in select scenarios, but in most cases, it remains less reliable than retraining from scratch
24,756
null
null
[ -0.01503993570804596, -0.045380961149930954, 0.011870604939758778, 0.04663458839058876, 0.04488121345639229, -0.004846206866204739, 0.02543555572628975, 0.030193796381354332, -0.01370844617486, -0.025136856362223625, -0.00043903509504161775, 0.01841522939503193, -0.0636519119143486, 0.0039...
HardcoreLogic: Challenging Large Reasoning Models with Long-tail Logic Puzzle Games
https://openreview.net/forum?id=8USxc43D3I
[ "Jingcong Liang", "Shijun Wan", "Xuehai Wu", "Yitong Li", "Qianglong Chen", "Duyu Tang", "Siyuan Wang", "zhongyu wei" ]
Poster
datasets and benchmarks
Large Reasoning Models (LRMs) have demonstrated impressive performance on complex tasks, including logical puzzle games that require deriving solutions satisfying all constraints. However, whether they can flexibly apply appropriate rules to varying conditions, particularly when faced with non-canonical game variants, ...
[ "long-tail benchmark", "logic puzzle games", "large reasoning model" ]
We propose HardcoreLogic, a logic puzzle game benchmark with non-canonical long-tail puzzles that evaluates the reasoning capability robustness of LLM/LRMs.
24,750
2510.12563
title_snapshot
[ -0.04810129478573799, -0.020051591098308563, -0.014043619856238365, 0.04855893924832344, 0.05568750947713852, -0.01853336952626705, 0.00014903576811775565, -0.0059351385571062565, -0.03761017695069313, -0.009540143422782421, -0.022495316341519356, 0.029473869130015373, -0.0521843321621418, ...
Efficient Credal Prediction through Decalibration
https://openreview.net/forum?id=BqOmsYIe7M
[ "Paul Hofman", "Timo Löhr", "Maximilian Muschalik", "Yusuf Sale", "Eyke Hüllermeier" ]
Poster
probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)
A reliable representation of uncertainty is essential for the application of modern machine learning methods in safety-critical settings. In this regard, the use of credal sets (i.e., convex sets of probability distributions) has recently been proposed as a suitable approach to representing epistemic uncertainty. Howev...
[ "efficient uncertainty representation", "credal sets", "relative likelihood" ]
Efficient credal prediction based on plausible probability intervals for computationally complex models (e.g. TabPFN, CLIP,…).
24,743
2603.08495
title_snapshot
[ 0.0026885035913437605, 0.0451848991215229, -0.01534857228398323, 0.0642302930355072, 0.03919399157166481, 0.04519713670015335, 0.026891330257058144, 0.018119443207979202, -0.042108193039894104, -0.03499766066670418, -0.04125284031033516, 0.0546211376786232, -0.04717972129583359, -0.0076410...
On the Lipschitz Continuity of Set Aggregation Functions and Neural Networks for Sets
https://openreview.net/forum?id=sPRK6XefjY
[ "Giannis Nikolentzos", "Konstantinos Skianis" ]
Poster
other topics in machine learning (i.e., none of the above)
The Lipschitz constant of a neural network is connected to several important properties of the network such as its robustness and generalization. It is thus useful in many settings to estimate the Lipschitz constant of a model. Prior work has focused mainly on estimating the Lipschitz constant of multi-layer perceptron...
[ "set aggregation functions", "Lipschitz continuity", "stability" ]
null
24,739
2505.24403
title_snapshot
[ -0.035951267927885056, 0.0007217896054498851, 0.007225791923701763, 0.03739893063902855, 0.02562687173485756, 0.04056625813245773, 0.019455229863524437, 0.009477430023252964, -0.017238300293684006, -0.04792547598481178, -0.013071928173303604, 0.037356846034526825, -0.09436959028244019, -0....
Let's Think in Two Steps: Mitigating Agreement Bias in MLLMs with Self-Grounded Verification
https://openreview.net/forum?id=rwo7bVlnzo
[ "Moises Andrade", "Joonhyuk Cha", "Brandon Ho", "Vriksha Srihari", "Karmesh Yadav", "Zsolt Kira" ]
Poster
foundation or frontier models, including LLMs
Verifiers—functions assigning rewards to agent behavior—have been key to AI progress in domains such as math, code, and games. However, extending these gains to domains without clear-cut success criteria (e.g., computer use) remains a challenge: while humans can recognize desired outcomes, translating this intuition in...
[ "Verifiers", "Verification", "Digital Agents", "Web Agents", "GUI Agents", "Robotics", "Large Language Models", "Test Time Scaling", "WebArena", "OSWorld", "Reward Models", "open-endedness", "LLMs-as-judges", "Vision Language Models" ]
null
24,728
2507.11662
title_snapshot
[ 0.017205070704221725, -0.02157825231552124, 0.0017738169990479946, 0.031916916370391846, 0.039375513792037964, -0.015432706102728844, 0.038965798914432526, -0.007982881739735603, -0.017362123355269432, -0.021854838356375694, -0.01255024503916502, 0.03477338328957558, -0.07150138169527054, ...
Communication-Efficient Decentralized Optimization via Double-Communication Symmetric ADMM
https://openreview.net/forum?id=HZYuyNkBdD
[ "Jinrui Huang", "Xueqin Wang", "Dong Liu", "Jingguo Lan", "Runxiong Wu" ]
Poster
optimization
This paper focuses on decentralized composite optimization over networks without a central coordinator. We propose a novel decentralized Symmetric ADMM algorithm that incorporates multiple communication rounds within each iteration, derived from a new constraint formulation that enables information exchange beyond imme...
[ "Decentralized Optimization", "Symmetric ADMM", "Multi-Communication" ]
null
24,719
2511.05283
title_snapshot
[ -0.022330455482006073, -0.03733555227518082, 0.0011629468062892556, 0.028278255835175514, 0.011413482949137688, 0.05389740318059921, 0.03079657256603241, -0.013459865935146809, -0.021811340004205704, -0.06343793869018555, 0.0033223354257643223, -0.01461104117333889, -0.05683700367808342, 0...
Meta-RL Induces Exploration in Language Agents
https://openreview.net/forum?id=4GiBscHW1k
[ "Yulun Jiang", "Liangze Jiang", "Damien Teney", "Michael Moor", "Maria Brbic" ]
Poster
foundation or frontier models, including LLMs
Reinforcement learning (RL) has enabled the training of Large Language Model (LLM) agents to interact with the environment and to solve multi-turn longhorizon tasks. However, the RL-trained agents often struggle in tasks that require active exploration and fail to efficiently adapt from trial-and-error experiences. In ...
[ "Large Language Model", "Agent", "Reinforcement Learning", "Meta Learning" ]
null
24,714
2512.16848
title_snapshot
[ -0.018436869606375694, -0.024888422340154648, 0.004068220499902964, 0.028370970860123634, 0.03326896205544472, 0.022598523646593094, 0.039962247014045715, 0.00917487870901823, -0.05082686245441437, 0.017177987843751907, -0.04180452227592468, 0.0532337911427021, -0.06435676664113998, -0.020...
Micro-Macro Retrieval: Reducing Long-Form Hallucination in Large Language Models
https://openreview.net/forum?id=ABdgMoJhlO
[ "Yujie Feng", "Jian Li", "Zhihan Zhou", "Pengfei Xu", "Yujia Zhang", "xiaoyu li", "Xiaohui Zhou", "Alan Zhao", "Xi Chen", "Xiao-Ming Wu" ]
Poster
generative models
Large Language Models (LLMs) achieve impressive performance across many tasks but remain prone to hallucination, especially in long-form generation where redundant retrieved contexts and lengthy reasoning chains amplify factual errors. Recent studies highlight a critical phenomenon: the closer key information appears t...
[ "Hallucination", "Long-form Hallucination", "Large Language Models" ]
null
24,705
2605.28828
title_snapshot
[ -0.022369325160980225, 0.003760579042136669, 0.006787377409636974, 0.042755164206027985, 0.02385970950126648, -0.0066729262471199036, 0.014356285333633423, 0.0012111278483644128, -0.028518175706267357, 0.004016685765236616, -0.02688218280673027, 0.038479454815387726, -0.06199231371283531, ...
Do We Really Need Permutations? Impact of Model Width on Linear Mode Connectivity
https://openreview.net/forum?id=ll8GLAic7q
[ "Akira Ito", "Masanori Yamada", "Daiki Chijiwa", "Atsutoshi Kumagai" ]
Poster
other topics in machine learning (i.e., none of the above)
Recently, Ainsworth et al. empirically demonstrated that, given two independently trained models, applying a parameter permutation that preserves the input–output behavior allows the two models to be connected by a low-loss linear path. When such a path exists, the models are said to achieve linear mode connectivity (L...
[ "deep learning", "linear mode connectivity", "permutation symmetries" ]
Width expansion probably facilitates linear mode connectivity without permutations.
24,703
2510.08023
title_snapshot
[ -0.010888950899243355, -0.009457866661250591, -0.027282241731882095, 0.026749208569526672, 0.05882234498858452, 0.014655107632279396, 0.018505258485674858, 0.013685151003301144, -0.02076253853738308, -0.056436289101839066, 0.01602794975042343, 0.007863101549446583, -0.08292773365974426, -0...
Enhancing Visual Token Representations for Video Large Language Models via Training-free Spatial-Temporal Pooling and Gridding
https://openreview.net/forum?id=MZi9SYPVz5
[ "Bingjun Luo", "Tony Wang", "Hanqi Chen", "Xinpeng Ding" ]
Poster
applications to computer vision, audio, language, and other modalities
Recent advances in Multimodal Large Language Models (MLLMs) have significantly advanced video understanding tasks, yet challenges remain in efficiently compressing visual tokens while preserving spatiotemporal interactions. Existing methods, such as LLaVA family, utilize simplistic pooling or interpolation techniques t...
[ "Visual Token Representation", "Video Understanding", "Multimodal Large Language Models" ]
Our training-free method, ST-GridPool, boosts Video LLM performance and efficiency by intelligently compressing visual tokens based on their spatiotemporal importance.
24,698
2605.22078
title_snapshot
[ 0.030180543661117554, -0.018915455788373947, 0.018321698531508446, 0.03292037174105644, 0.018426433205604553, 0.023866888135671616, 0.002282237634062767, 0.030264224857091904, -0.040959883481264114, -0.02083154022693634, -0.009689482860267162, -0.0350283682346344, -0.056802745908498764, 0....
ProxyAttn: Guided Sparse Attention via Representative Heads
https://openreview.net/forum?id=m3HXHQYmZu
[ "Yixuan Wang", "Huang He", "Siqi Bao", "Hua Wu", "Haifeng Wang", "Qingfu Zhu", "Wanxiang Che" ]
Poster
applications to computer vision, audio, language, and other modalities
The quadratic complexity of attention mechanisms limits the efficiency of Large Language Models (LLMs) on long-text tasks. Recently, methods that dynamically estimate block importance have enabled efficient block sparse attention, leading to significant acceleration in long-text pre-filling of LLMs. However, their bloc...
[ "Efficient LLM", "Sparse Attention" ]
null
24,686
2509.24745
title_snapshot
[ 0.023779094219207764, -0.01051238551735878, -0.0020848303101956844, 0.02164514921605587, 0.015201972797513008, 0.01923040673136711, 0.03189924359321594, 0.013745144940912724, -0.04721969738602638, -0.014820990152657032, -0.009330150671303272, 0.012361194007098675, -0.0577130950987339, 0.01...
LightMem: Lightweight and Efficient Memory-Augmented Generation
https://openreview.net/forum?id=dyJ0GWpjJB
[ "Jizhan Fang", "Xinle Deng", "Haoming Xu", "Ziyan Jiang", "Yuqi Tang", "Ziwen Xu", "Shumin Deng", "Yunzhi Yao", "Mengru Wang", "Shuofei Qiao", "Huajun Chen", "Ningyu Zhang" ]
Poster
foundation or frontier models, including LLMs
Despite their remarkable capabilities, Large Language Model (LLM) struggle to effectively leverage historical interaction information in dynamic and complex environments. Memory systems enable LLMs to move beyond stateless interactions by introducing persistent information storage, retrieval, and utilization mechanisms...
[ "large language model", "LLM memory" ]
null
24,681
2510.18866
title_snapshot
[ -0.017361976206302643, 0.0014868183061480522, -0.01035385299474001, 0.023269619792699814, 0.02868264727294445, 0.006323268171399832, -0.011240948922932148, 0.03143556788563728, -0.05274743214249611, -0.026760660111904144, -0.0248723067343235, 0.01629679836332798, -0.07771718502044678, 0.01...
TRAC: Tensor-Train based Across-layer Compression for Parameter-Efficient Fine-Tuning
https://openreview.net/forum?id=tz5yPWZp9W
[ "Bangguo Ye", "Yuanwei Zhang", "Xiaoqun Zhang" ]
Poster
transfer learning, meta learning, and lifelong learning
Fine-tuning large pre-trained models under resource constraints remains challenging due to the massive number of parameters involved. Existing parameter-efficient tuning methods, such as low-rank adaptation (LoRA) and its variants, rely heavily on matrix factorization and often struggle in extremely low-parameter regim...
[ "Parameter-efficient fine-tuning", "Low-rank adaptation", "Tensor decomposition" ]
We propose TRAC, a tensor-based extension of LoRA that enables extremely parameter-efficient fine-tuning while preserving strong model performance.
24,680
null
null
[ -0.009578075259923935, -0.03651167452335358, -0.02350642904639244, 0.035629525780677795, 0.024771813303232193, 0.047541942447423935, -0.013012549839913845, 0.009515456855297089, -0.014491318725049496, -0.039399877190589905, -0.03663893789052963, 0.014565289951860905, -0.05908765271306038, ...
ST-SimDiff: Balancing Spatiotemporal Similarity and Difference for Efficient Video Understanding with MLLMs
https://openreview.net/forum?id=he8kYNcoMA
[ "Bingjun Luo", "Tony Wang", "Chaoqi Chen", "Xinpeng Ding" ]
Poster
applications to computer vision, audio, language, and other modalities
Multimodal Large Language Models (MLLMs) face significant computational overhead when processing long videos due to the massive number of visual tokens required. To improve efficiency, existing methods primarily reduce redundancy by pruning or merging tokens based on importance or similarity. However, these approaches ...
[ "Video Understanding", "Visual Token Reduction", "Multimodal Large Language Models" ]
We propose a training-free method that builds a spatio-temporal graph to efficiently select video tokens by incorporating similarity for redundancy reduction and difference for key event detection.
24,671
2605.22158
title_snapshot
[ 0.03722840175032616, -0.045922283083200455, 0.019811438396573067, 0.043160103261470795, 0.015185633674263954, 0.03544585779309273, 0.028692495077848434, 0.030908558517694473, -0.02562764286994934, -0.02992229349911213, -0.000975057715550065, -0.009022744372487068, -0.06318306177854538, 0.0...
Flow2GAN: Hybrid Flow Matching and GAN with Multi-Resolution Network for Few-step High-Fidelity Audio Generation
https://openreview.net/forum?id=5eTpRIULtb
[ "Zengwei Yao", "Wei Kang", "Han Zhu", "Liyong Guo", "Lingxuan Ye", "Fangjun Kuang", "Weiji Zhuang", "Zhaoqing Li", "Zhifeng Han", "Long Lin", "Daniel Povey" ]
Poster
applications to computer vision, audio, language, and other modalities
Existing dominant methods for audio generation include Generative Adversarial Networks (GANs) and diffusion-based methods like Flow Matching. GANs suffer from slow convergence during training, while diffusion methods require multi-step inference that introduces considerable computational overhead. In this work, we intr...
[ "Flow2GAN", "audio generation", "Flow Matching", "GAN", "multi-resolution" ]
null
24,663
2512.23278
title_snapshot
[ 0.00016709939518477768, -0.023161953315138817, -0.010714994743466377, 0.024606017395853996, 0.013520551845431328, 0.040834274142980576, 0.032392892986536026, -0.016066664829850197, 0.0005802787491120398, -0.07377506047487259, 0.009698010049760342, 0.0020673777908086777, -0.045579515397548676...
ContextIF: Enhancing Instruction-Following through Context Reward
https://openreview.net/forum?id=IuscGSmfEf
[ "Yule Zhong", "Jiacheng Yao", "Guoxiu He" ]
Poster
reinforcement learning
While supervised fine-tuning (SFT) and preference learning (PL) are widely used to enhance the instruction-following ability of Large Language Models (LLMs), they often struggle to generalize to novel or complex instructions and may compromise the models' general capabilities. In-Context Learning (ICL) emerges as a pro...
[ "Large Language Models", "Instruction-Following", "Reinforcement Learning", "In-Context Learning" ]
Enhancing instruction-following of LLMs by autonomously generating optimal contexts via reinforcement learning with comprehensive context rewards.
24,662
null
null
[ -0.014301237650215626, -0.0056355055421590805, -0.01726914756000042, 0.027487315237522125, 0.01898503489792347, 0.026343127712607384, 0.030120955780148506, 0.028167814016342163, -0.025738371536135674, -0.01595018059015274, -0.03106284886598587, 0.06389158964157104, -0.04264311119914055, -0...
Use the Online Network If You Can: Towards Fast and Stable Reinforcement Learning
https://openreview.net/forum?id=rFLuaG9Yq6
[ "Ahmed Hendawy", "Henrik Metternich", "Théo Vincent", "Mahdi Kallel", "Jan Peters", "Carlo D'Eramo" ]
Poster
reinforcement learning
The use of target networks is a popular approach for estimating value functions in deep Reinforcement Learning (RL). While effective, the target network remains a compromise solution that preserves stability at the cost of slowly moving targets, thus delaying learning. Conversely, using the online network as a bootstra...
[ "deep reinforcement learning", "q-learning", "actor-critic", "function approximation" ]
MINTO is a simple, yet effective target bootstrapping method for temporal-difference RL that enables faster, more stable learning and consistently improves performance across algorithms and benchmarks.
24,658
2510.02590
title_snapshot
[ -0.020356759428977966, -0.017966220155358315, -0.0020734204445034266, 0.030396616086363792, 0.044726189225912094, 0.034330058842897415, -0.0019332244992256165, 0.014400476589798927, -0.03607672080397606, -0.035854220390319824, 0.004122718703001738, 0.013389918021857738, -0.08773341029882431,...
Efficient Prediction of Large Protein Complexes via Subunit-Guided Hierarchical Refinement
https://openreview.net/forum?id=0G8Cq9z2Hp
[ "Chixiang Lu", "Yunhua Zhong", "Shikang Liang", "Xiaojuan Qi", "Haibo Jiang" ]
Poster
applications to physical sciences (physics, chemistry, biology, etc.)
State-of-the-art protein structure predictors have revolutionized structural biology, yet quadratic memory growth with token length makes end-to-end inference impractical for large complexes beyond a few thousand tokens. We introduce HierAFold, a hierarchical pipeline that exploits the modularity of large complexes via...
[ "Protein complex structure prediction", "AlphaFold3", "complex modularity" ]
We introduce HierAFold, a hierarchical pipeline that exploits the modularity of large complexes via PAE-guided (Predicted Aligned Error) subunit decomposition, targeted interface-aware refinement, and confidence-weighted assembly.
24,657
null
null
[ -0.041612643748521805, 0.0038259366992861032, -0.013729212805628777, -0.005486982874572277, 0.010410821065306664, 0.0076104882173240185, 0.02006274275481701, -0.00955736543983221, 0.0012775335926562548, -0.024977020919322968, 0.040662843734025955, -0.004953397903591394, -0.06754270195960999,...
FlowSearcher: Synthesizing Memory-Guided Agentic Workflows for Web Information Seeking
https://openreview.net/forum?id=34v7DVz2l0
[ "Keyi Xiang", "Zeyu Feng", "Zhuoyi Lin", "Yueming Lyu", "Shi Boyuan", "Yew-Soon Ong", "Ivor Tsang", "Haiyan Yin" ]
Poster
applications to computer vision, audio, language, and other modalities
Web search is a cornerstone for deep research agents, enabling them to acquire and reason over knowledge beyond static corpora. Yet most existing systems rely on ReAct-style tool chains with rigid, linear workflows, hindering their ability to adapt to diverse query types and tool-use strategies. We introduce **FlowSea...
[ "Large Language Model Reasoning", "Structured Planning", "Agentic Workflow" ]
FlowSearcher is the first to reframe web search as query-specific agentic workflow synthesis with structured planning and hierarchical memory.
24,655
null
null
[ -0.01691463403403759, -0.047213684767484665, -0.013002783060073853, 0.03367561101913452, 0.042620301246643066, 0.011652650311589241, 0.02738446369767189, 0.028893180191516876, 0.007437472697347403, -0.049113016575574875, -0.030519157648086548, -0.004284956026822329, -0.044710878282785416, ...