ICLR 2026 Papers — Page 15
International Conference on Learning Representations · 5356 papers
Efficient Offline Reinforcement Learning via Peer-Influenced Constraint
Yujia Zhang (North University of China), Jiye Liang (Shanxi University)
Reinforcement LearningBenchmark
🎯 What it does: Proposes an offline reinforcement learning framework based on Peer Influence Constraint (PIC) and its integrated version (EPIC) to enhance policy generalization and efficiency.
Efficient Orthogonal Fine-Tuning with Principal Subspace Adaptation
Fei Wu (University of Exeter), Shiqiang Wang (University of Exeter)
Computational EfficiencyRepresentation LearningTransformerSupervised Fine-TuningImageText
🎯 What it does: Proposes the PSOFT method, performing orthogonal fine-tuning within the principal subspace of pre-trained models, balancing semantic preservation, expressiveness, and multidimensional efficiency.
Efficient Prediction of Large Protein Complexes via Subunit-Guided Hierarchical Refinement
Chixiang Lu (University of Hong Kong), Haibo Jiang (University of Hong Kong)
Protein Structure PredictionTransformerBiomedical Data
🎯 What it does: Propose HIERAFOLD, a hierarchical coarse-to-fine pipeline for predicting the structure of large protein complexes, which utilizes PAE-guided subunit segmentation, interface-aware refinement, and confidence-weighted assembly to significantly reduce memory usage and enable prediction of complexes with over 5k tokens.
Efficient Quantization of Mixture-of-Experts with Theoretical Generalization Guarantees
Mohammed Nowaz Rabbani Chowdhury (Rensselaer Polytechnic Institute), Meng Wang (Rensselaer Polytechnic Institute)
Computational EfficiencyTransformerMixture of ExpertsText
🎯 What it does: Proposed an expert-level mixed-precision quantization method based on router norm variations, which can assign different bit-widths to different experts in MoE models.
Efficient Reasoning with Balanced Thinking
Yulin Li (Harbin Institute of Technology), Zhuotao Tian (Harbin Institute of Technology)
Computational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose a training-free framework named REBALANCE that dynamically controls overthinking and underthinking in large reasoning models, achieving efficient balanced reasoning.
Efficient Regression-based Training of Normalizing Flows for Boltzmann Generators
Danyal Rehman, Joey Bose
GenerationData SynthesisDrug DiscoveryProtein Structure PredictionFlow-based ModelBiomedical Data
🎯 What it does: Propose REGFLOW, a regression-based training method that achieves high-quality sampling in single-step regularized flows while providing accurate likelihoods;
Efficient Reinforcement Learning by Guiding World Models with Non-Curated Data
Yi Zhao (Aalto University), Joni Pajarinen (Aalto University)
Robotic IntelligenceReinforcement LearningWorld ModelSequentialRetrieval-Augmented Generation
🎯 What it does: Pre-train a task-agnostic world model using unsupervised, mixed-quality, multi-body non-episodic offline data, and during online fine-tuning, reduce distribution shift through experience replay and execution guidance, significantly improving sample efficiency.
Efficient Resource-Constrained Training of Transformers via Subspace Optimization
Le-Trung Nguyen (Télécom Paris), Van-Tam Nguyen (Télécom Paris)
Computational EfficiencyTransformerSupervised Fine-TuningImageText
🎯 What it does: Proposed the Weight-Activation Subspace Iteration (WASI) method for efficiently training Transformer models on edge devices, compressing both weights and activations.
Efficient Sliced Wasserstein Distance Computation via Adaptive Bayesian Optimization
Manish Acharya (Vanderbilt University), David Hyde (Vanderbilt University)
OptimizationComputational EfficiencyImagePoint Cloud
🎯 What it does: This paper proposes Bayesian optimization-based projection direction selectors (BOSW, RBOSW, ABOSW, ARBOSW) to improve the estimation efficiency of sliced Wasserstein (SW) distance, which can be seamlessly integrated into existing SW optimization processes;
Efficient Submodular Maximization for Sums of Concave over Modular Functions
Yang Lv (Beijing University of Technology), Ruiqi Yang (Beijing University of Technology)
OptimizationGraph
🎯 What it does: This paper proposes a method utilizing accelerated approximate projected gradient ascent (AAPGA), combined with continuous relaxation, convex/concave extension, randomization, and pipelined rounding, to solve submodular maximization problems under cardinality, knapsack, and partition matroid constraints for 'additive concave modular' (SCM) functions;
Efficient Test-Time Scaling for Small Vision-Language Models
Mehmet Onurcan Kaya (Technical University of Denmark), Dim Papadopoulos (Technical University of Denmark)
Computational EfficiencyTransformerSupervised Fine-TuningVision Language ModelMultimodality
🎯 What it does: This paper proposes two efficient inference-time scaling methods for small vision-language models, namely Test-Time Augmentation (TTAug) and Test-Time Adaptation (TTAdapt).
Efficient Testing for Correlation Clustering: Improved Algorithms and Optimal Bounds
Chengyuan Deng (Rutgers University), Chen Wang (Rensselaer Polytechnic Institute)
OptimizationComputational EfficiencyGraph
🎯 What it does: This paper proposes an efficient attribute testing algorithm to determine whether a large complete labeled graph is clusterable (or k-clusterable, structurally balanced) and estimate clustering cost;
Efficient Turing Machine Simulation with Transformers
Qian Li (Shenzhen International Center For Industrial And Applied Mathematics), Yuyi Wang (CRRC Zhuzhou Institute)
Computational EfficiencyTransformerChain-of-Thought
🎯 What it does: This paper proves that constant-bit Transformers can simulate any (t(n), s(n))-bounded multitape Turing machine within the optimal O(s(n)) context window, using only O(s(n)^c) CoT steps.
Efficient Zero-shot Inpainting with Decoupled Diffusion Guidance
Badr MOUFAD, Jimmy Olsson (KTH Royal Institute of Technology)
RestorationDiffusion modelImage
🎯 What it does: This paper proposes a zero-shot image inpainting method called DING, which utilizes pre-trained diffusion models to perform fast and high-quality image restoration in the latent space;
Efficient-LVSM: Faster, Cheaper, and Better Large View Synthesis Model via Decoupled Co-Refinement Attention
Xiaosong Jia (Fudan University), Yu-Gang Jiang (Fudan University)
GenerationData SynthesisTransformerImage
🎯 What it does: Proposed an efficient dual-stream Transformer architecture, Efficient-LVSM, to achieve faster, more resource-efficient, and higher quality view synthesis;
Efficient-SAM2: Accelerating SAM2 with Object-Aware Visual Encoding and Memory Retrieval
Jing Zhang (Institute of Automation Chinese Academy of Sciences), Qingyi Gu (Institute of Automation Chinese Academy of Sciences)
SegmentationComputational EfficiencyTransformerVideo
🎯 What it does: Accelerate the post-training of SAM2 by introducing two modules: object-aware sparse window routing (SWR) and sparse memory retrieval (SMR), reducing redundant computations in the image encoder and memory attention.
EffiVMT: Video Motion Transfer via Efficient Spatial-Temporal Decoupled Finetuning
Yue Ma (Hong Kong University Of Science And Technology), Qifeng Chen (Hong Kong University Of Science And Technology)
GenerationTransformerSupervised Fine-TuningDiffusion modelVideoTextBenchmark
🎯 What it does: Fine-tune a pre-trained video Diffusion Transformer using the three-stage EffiVMT framework to achieve high-quality motion transfer, which can accurately replicate the motion of the reference video while preserving the visual appearance.
Egalitarian Gradient Descent: A Simple Approach to Accelerated Grokking
Ali Saheb Pasand (McGill University), Elvis Dohmatob (Concordia University)
OptimizationImage
🎯 What it does: Proposed Egalitarian Gradient Descent (EGD), a simple hyperparameter-free method that balances optimization speed along the principal gradient directions to accelerate the model's grokking process;
EGG-SR: Embedding Symbolic Equivalence into Symbolic Regression via Equality Graph
Nan Jiang (University of Texas - El Paso), Yexiang Xue (Purdue University)
Large Language ModelReinforcement LearningTabularBenchmarkPhysics Related
🎯 What it does: Propose a unified framework EGG-SR that embeds symbolic equivalence into symbolic regression algorithms, leveraging e-graphs to accelerate MCTS, DRL, and LLM methods.
Ego-Foresight: Self-supervised Learning of Agent-Aware Representations for Improved RL
Manuel Serra Nunes (Instituto Superior Técnico, Universidade de Lisboa), Jose Santos-Victor
Representation LearningRobotic IntelligenceConvolutional Neural NetworkRecurrent Neural NetworkReinforcement LearningAuto EncoderVideoSequential
🎯 What it does: Propose the Ego-Foresight method, which decouples agent information through self-supervised motion prediction and uses it as an auxiliary task for feature learning to improve sample efficiency in reinforcement learning.
EgoBrain: Synergizing Minds and Eyes For Human Action Understanding
Nie Lin (University of Tokyo), Dongsheng Li (Microsoft Research Asia)
RecognitionTransformerVision-Language-Action ModelVideoMultimodalityBiomedical DataBenchmark
🎯 What it does: Constructed the EgoBrain dataset, first to synchronously record first-person videos and 32-channel EEG from 40 subjects, covering 29 categories of daily activities; proposed the Brain-TIM model for cross-modal temporal fusion, achieving multimodal action recognition;
EgoDex: Learning Dexterous Manipulation from Large-Scale Egocentric Video
Ryan Hoque (Apple), Jian Zhang (Apple)
Pose EstimationRobotic IntelligenceTransformerDiffusion modelFlow-based ModelVideoBenchmark
🎯 What it does: Constructed the largest egocentric viewpoint grasping dataset EgoDex, and evaluated imitation learning methods for hand trajectory prediction based on it.
EgoHandICL: Egocentric 3D Hand Reconstruction with In-Context Learning
Binzhu Xie (Chinese University of Hong Kong), Pheng-Ann Heng (Khalifa University)
Pose EstimationRetrievalTransformerVision Language ModelAuto EncoderImageVideoMultimodality
🎯 What it does: Propose the EgoHandICL framework to achieve adaptive 3D hand reconstruction, leveraging VLM for retrieving example templates, ICL tokenizer for multimodal context tokenization, and MAE-based architecture;
EgoNight: Towards Egocentric Vision Understanding at Night with a Challenging Benchmark
Deheng Zhang (INSAIT, Sofia University St Kliment Ohridski), Danda Pani Paudel (INSAIT, Sofia University St Kliment Ohridski)
RecognitionData SynthesisDepth EstimationRetrievalTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: This paper proposes the EgoNight benchmark kit, covering nighttime egocentric visual tasks, including VQA, day-night correspondence retrieval, and depth estimation, and generates 3,658 high-quality QA pairs through human-machine collaborative automatic annotation.
EgoTwin: Dreaming Body and View in First Person
Jingqiao Xiu (National University of Singapore), Ziwei Liu (Shanghai AI Laboratory)
GenerationData SynthesisTransformerDiffusion modelVideoTextMultimodality
🎯 What it does: Propose a diffusion-based framework called EgoTwin that can synchronously generate first-person videos and human motion.
EgoWorld: Translating Exocentric View to Egocentric View using Rich Exocentric Observations
Junho Park (LG Electronics), Taein Kwon (University of Oxford)
Image TranslationGenerationData SynthesisPose EstimationDepth EstimationTransformerVision Language ModelDiffusion modelAuto EncoderImageTextMultimodalityPoint Cloud
🎯 What it does: Synthesize high-quality first-person perspective images through a single appearance image by leveraging three multimodal observations: point cloud, 3D hand pose, and text description, using a two-stage end-to-end pipeline (first generating sparse egocentric RGB and pose, then completing the synthesis with a diffusion model).
Eigen-Agent: Adaptive Multi-Agent Scientific Reasoning with Monitor-Based RAG
Xiangru Tang (Yale University), Di Jin (Eigen AI)
Agentic AITextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose EIGEN-AGENT, combining monitor-style retrieval and hierarchical schemes to improve multi-agent scientific reasoning, significantly enhancing the accuracy on HLE Bio/Chem.
EigenBench: A Comparative Behavioral Measure of Value Alignment
Jonathn Chang (Cornell University), Lionel Levine (Cornell University)
Explainability and InterpretabilityLarge Language ModelTextBenchmark
🎯 What it does: Propose the EigenBench method, which quantifies the average alignment of language models with a given constitution (value system) using model peer review and the EigenTrust algorithm.
EigenScore: OOD Detection using Posterior Covariance in Diffusion Models
Shirin Shoushtari (Washington University in St. Louis), Ulugbek S. Kamilov (University of Wisconsin-Madison)
Anomaly DetectionDiffusion modelScore-based ModelImage
🎯 What it does: Propose the EigenScore method, which uses the eigen-spectrum of the posterior covariance for unsupervised OOD detection.
Einstein Fields: A Neural Perspective To Computational General Relativity
Sandeep Suresh Cranganore (JKU Linz), Johannes Brandstetter (JKU Linz)
MeshPhysics Related
🎯 What it does: Neurally compress the metric tensor in 4D general relativity simulations, constructing the EinFields model, which can reconstruct the complete spacetime geometry and perform differentiation with extremely low storage requirements.
EIP: Weighted Ranking of LLMs by Quantifying Question Difficulty
Xingjian Hu (Lehigh University), Lichao Sun (Lehigh University)
Computational EfficiencyLarge Language ModelTextBenchmark
🎯 What it does: Proposed and validated the Empirical Interaction Propagation (EIP) framework for simultaneously estimating problem difficulty and model capability in large language model assessments;
Elastic Optimal Transport: Theory, Application, and Empirical Evaluation
Pei Yang (South China University of Technology), Qi Tan (South China Normal University)
Domain AdaptationOptimizationImageBenchmark
🎯 What it does: Propose the Elastic Optimal Transport (ELOT) framework, which can adaptively transport partial mass between probability distributions and achieve more robust distribution matching in domain adaptation tasks.
ELEPHANT: Measuring and understanding social sycophancy in LLMs
Myra Cheng (Stanford University), Dan Jurafsky (Stanford University)
TransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextBenchmark
🎯 What it does: This work introduces the concept of 'social sycophancy,' viewing excessive agreement in LLMs as a preservation of user face (self-image), and designs the ELEPHANT benchmark to systematically quantify this phenomenon.
Eliciting Harmful Capabilities by Fine-Tuning on Safeguarded Outputs
Jackson Kaunismaa (MATS), Erik Jones (Anthropic)
Adversarial AttackLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Propose to fine-tune open-source models using harmless outputs generated by a safety-frontend model, thereby enhancing their ability to perform harmful tasks;
Eliciting Numerical Predictive Distributions of LLMs Without Auto-Regression
Julianna Piskorz (University of Cambridge), Mihaela van der Schaar (University of Cambridge)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningTransformerLarge Language ModelTime Series
🎯 What it does: Investigate how to recover numerical prediction point estimates (mean, median, greedy value) and their uncertainty distributions directly from hidden states by probing the internal representations of large language models (LLMs), without requiring autoregressive sampling.
Eliminating Inductive Bias in Reward Models with Information-Theoretic Guidance
Zhuo Li (Alibaba), guanjunjiang
Reinforcement Learning from Human FeedbackTransformerReinforcement LearningText
🎯 What it does: Studied a reward model debiasing method called DIR, aiming to eliminate inductive bias caused by low-quality human preference data in RLHF training (e.g., response length, flattery tone, formatting, etc.).
Eliminating VAE for Fast and High-Resolution Generative Detail Restoration
Yan Wang (ByteDance), Li zhang
RestorationGenerationSuper ResolutionDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: Propose a high-resolution image restoration method that completely removes the VAE from diffusion models and directly performs generation in the pixel space;
ELLMob: Event-Driven Human Mobility Generation with Self-Aligned LLM Framework
Yusong Wang (Institute of Science Tokyo), Renhe Jiang (University of Tokyo)
GenerationData SynthesisTransformerLarge Language ModelTextTime Series
🎯 What it does: Propose an event-driven human mobility generation framework based on large language models, named ELLMob, and construct the first mobile dataset containing three major events, including typhoons, COVID-19, and the Olympics.
ELMUR: External Layer Memory with Update/Rewrite for Long-Horizon RL Problems
Egor Cherepanov (AXXX), Aleksandr Panov (AXXX)
TransformerReinforcement LearningMixture of ExpertsBenchmark
🎯 What it does: Propose a new Transformer architecture ELMUR, achieving memory and backtracking of long-term dependencies by introducing external updatable memory at each layer and using LRU write-back mechanism;
ELViS: Efficient Visual Similarity from Local Descriptors that Generalizes Across Domains
Pavel Suma (Czech Technical University in Prague), Giorgos Tolias (Czech Technical University in Prague)
RetrievalDomain AdaptationTransformerImage
🎯 What it does: Propose ELViS, a lightweight image-to-image similarity model based on a local feature similarity matrix, for re-ranking in retrieval tasks.
Embedding-Based Context-Aware Reranker
Ye Yuan (McGill University), Siqi Liu (RBC Borealis)
RetrievalComputational EfficiencyRepresentation LearningTransformerContrastive LearningTextRetrieval-Augmented Generation
🎯 What it does: Proposed a lightweight, embedding-based, context-aware re-ranker called EBCAR, specifically designed for re-ranking tasks in retrieval-augmented generation (RAG);
Embodied Agents Meet Personalization: Investigating Challenges and Solutions Through the Lens of Memory Utilization
Taeyoon Kwon (Yonsei University), Jinyoung Yeo (Yonsei University)
Safty and PrivacyRobotic IntelligenceTransformerLarge Language ModelPrompt EngineeringTextGraphBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose the MEMENTO framework to evaluate the memory utilization capability of LLM-driven embodied agents in personalized assistance tasks, revealing the agents' weaknesses in retrieval and memory coordination through two-phase experiments (single memory and joint memory).
Embodied Navigation Foundation Model
Jiazhao Zhang (Peking University), He Wang (Peking University)
Autonomous DrivingRobotic IntelligenceTransformerLarge Language ModelVision Language ModelImageVideoTextMultimodality
🎯 What it does: Proposed a cross-task and cross-implementation navigation foundation model called NavFoM, capable of handling visual language navigation for various robots and tasks.
Embodied-R1: Reinforced Embodied Reasoning for General Robotic Manipulation
Yifu Yuan (College of Intelligence and Computing, Tianjin University), Jianye HAO
Robotic IntelligenceTransformerLarge Language ModelReinforcement LearningVision-Language-Action ModelDiffusion modelImageTextMultimodalityPoint Cloud
🎯 What it does: Proposed and trained Embodied-R1, an embodied reasoning vision-language model that utilizes pointing representations to perform localization, spatial relation reasoning, functional annotation, and visual trajectory generation across four pointing capabilities (REG, RRG, OFG, VTG), achieving zero-shot robot control through these pointing results.
Embracing Discrete Search: A Reasonable Approach to Causal Structure Learning
Marcel Wienöbst, Sebastian Weichwald (University College Dublin)
OptimizationComputational EfficiencyGraph
🎯 What it does: Propose the FLOP algorithm that uses discrete search and incremental Cholesky updates to efficiently optimize the BIC score of linear models with noise, quickly learning the DAG structure.
EMBridge: Enhancing Gesture Generalization from EMG Signals Through Cross-modal Representation Learning
Wenhui Cui (Apple), Behrooz Mahasseni (Apple)
ClassificationRepresentation LearningTransformerAuto EncoderContrastive LearningMultimodalityBiomedical Data
🎯 What it does: Propose the EMBridge framework, which aligns low-quality sEMG signals with high-quality hand pose data through cross-modal alignment, enhancing EMG representation quality and enabling zero-shot gesture classification.
Emergence of Spatial Representation in an Actor-Critic Agent with Hippocampus-Inspired Sequence Generator
Xiao-Xiong Lin (University of Freiburg), Christian Leibold (University of Freiburg)
Computational EfficiencyRepresentation LearningConvolutional Neural NetworkRecurrent Neural NetworkReinforcement LearningImage
🎯 What it does: Developed a shift-register RNN inspired by the hippocampus CA3, combined with sparse DG input and actor-critic learning, achieving self-centered perception-driven navigation in the visual-based DeepMind Lab continuous maze;
Emergence of Superposition: Unveiling the Training Dynamics of Chain of Continuous Thought
Hanlin Zhu (University Of California Berkeley), Yuandong Tian (Meta Ai)
Explainability and InterpretabilityTransformerLarge Language ModelGraphChain-of-Thought
🎯 What it does: This paper investigates how continuous Chain-of-Thought (CoT) naturally generates a superposition mechanism during training through theoretical analysis and experimental verification, and reveals the training dynamics of this mechanism.
Emergent Coordination in Multi-Agent Language Models
Christoph Riedl (Northeastern University)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper designs a non-communicative group guessing game, quantitatively evaluates the emergence, collaboration, and identity differences in multi-agent LLM systems using information theory (PID, TDMI) and mixed-effects models, and investigates how prompts (Plain, Persona, ToM) can control systems to transition from disordered collections to higher-order collaborative collectives.
Emergent Dexterity Via Diverse Resets and Large-Scale Reinforcement Learning
Patrick Yin (University of Washington), Abhishek Gupta (University of Washington)
Robotic IntelligenceReinforcement LearningImageSequential
🎯 What it does: Propose the OmniReset framework, which generates diverse reset distributions (including reaching, near-object, stable grasp, and near-goal states) programmatically to provide rich initial conditions for large-scale parallel RL (PPO), enabling complex dexterous manipulation tasks with only a single generic reward, no curriculum, and no example demonstrations;
Emergent Discrete Controller Modules for Symbolic Planning in Transformers
S M Rafiuddin (Oklahoma State University), Muntaha Nujat Khan (Oklahoma State University)
Explainability and InterpretabilityAI Code AssistantTransformerTextSequential
🎯 What it does: Embed a learnable discrete controller module within Transformer layers, utilizing program primitives such as ASSIGN, ADD, COMPARE, BRANCH and state registers, enabling the model to perform symbolic planning, variable updates, and conditional branching between self-attention and feed-forward networks, thereby significantly enhancing extrapolation capabilities for algorithmic tasks.
Emergent Hierarchical Reasoning in LLMs through Reinforcement Learning
Haozhe Wang (Hong Kong University of Science and Technology), Wenhu Chen (Hong Kong University of Science and Technology)
TransformerLarge Language ModelReinforcement LearningTextMultimodality
🎯 What it does: Through a fine-grained analysis of the RL training process, it is found that the reasoning ability of LLMs is driven by a two-phase dynamic: first mastering low-level execution skills and then breaking through high-level strategic planning, and a hierarchical credit assignment method (HICRA) targeting high-level planning is proposed;
Emergent Misalignment is Easy, Narrow Misalignment is Hard
Anna Soligo (Imperial College London), Neel Nanda (Imperial College London)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Investigated the 'emergent misalignment' phenomenon in large language models after fine-tuning on fine-grained harmful data, and proposed linear direction representations along with their monitoring and mitigation methods.
EMFuse: Energy-based Model Fusion for Decision Making
Kejie He (Nanjing University), Yang Yu (Nanjing University)
TransformerLarge Language ModelReinforcement LearningTextSequential
🎯 What it does: Proposed a unified energy model fusion framework (EMFuse) that merges models through energy addition in two decision tasks: direct strategy fusion and dynamics model fusion, and designed EMSelect strategy selection and ADETM energy dynamics model to achieve efficient uncertainty estimation and fusion.
EmoPrefer: Can Large Language Models Understand Human Emotion Preferences?
Zheng Lian (Tongji University), Jianhua Tao (Tsinghua University)
RecognitionTransformerLarge Language ModelPrompt EngineeringMixture of ExpertsVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: Constructed the first emotional preference dataset EmoPrefer-Data and proposed EmoPrefer-Bench, establishing a benchmark for evaluating the emotional preference judgment of multimodal large language models (MLLM) in descriptive emotional recognition (DMER).
EmotionHallucer: Evaluating Emotion Hallucinations in Multimodal Large Language Models
Bohao Xing (Lappeenranta-Lahti University of Technology), Heikki Kälviäinen
Explainability and InterpretabilityLarge Language ModelTextMultimodalityBenchmarkAudio
🎯 What it does: Proposed the EmotionHallucer benchmark to assess hallucinations in multimodal large language models for emotion understanding.
Emotions Where Art Thou: Understanding and Characterizing the Emotional Latent Space of Large Language Models
Benjamin Reichman (Georgia Institute of Technology), Larry Heck (Georgia Institute of Technology)
ClassificationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Investigate the internal emotional representations in large language models, discovering that hidden layers contain a low-dimensional emotional subspace and that emotional manipulation can be achieved through learning modules.
EmotionThinker: Prosody-Aware Reinforcement Learning for Explainable Speech Emotion Reasoning
Dingdong WANG, Helen M. Meng (The Chinese University of Hong Kong)
ClassificationRecognitionExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextMultimodalityChain-of-ThoughtAudio
🎯 What it does: Propose EmotionThinker, a reinforcement learning-based speech emotion recognition model capable of generating interpretable emotional judgments and reasoning processes.
Empowering Efficiency and Efficacy in WebAgent via Enabling Info-Rich Seeking
Zhengwei Tao (Peking University), Zhiqiang Gao (Southeast University)
RetrievalComputational EfficiencyLarge Language ModelAgentic AITabular
🎯 What it does: Developed a WebLeaper framework that trains LLMs for efficient information seeking through entity-dense tree-structured task synthesis and information-oriented trajectory filtering.
Empowering LLM Tool Invocation with Tool-call Reward Model
Da Ma (Shanghai Jiao Tong University), Lu Chen (Shanghai Jiao Tong University)
AI Code AssistantReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Proposed the Tool Call Reward Model (TRM), providing fine-grained rewards for large language models when using external tools, and combined PPO and GRPO reinforcement learning to enhance performance in search and code generation tasks.
Empowering Multi-Robot Cooperation via Sequential World Models
Zijie Zhao (School Of Artificial Intelligence University Of Chinese Academy Of Sciences), Dongbin Zhao
Robotic IntelligenceReinforcement LearningWorld ModelSequential
🎯 What it does: Propose the SeqWM framework, introducing the sequential paradigm into model-based reinforcement learning for multi-robot systems. It employs independent autoregressive world models and sequential planning to achieve intent sharing and collaborative control, with performance validated on Bi-DexHands and Multi-Quad environments as well as real-world quadruped robots.
Empowering Small VLMs to Think with Dynamic Memorization and Exploration
Jiazhen Liu (Hong Kong University of Science and Technology), Long Chen (Hong Kong University of Science and Technology)
Supervised Fine-TuningReinforcement LearningPrompt EngineeringVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: Proposes a dynamic memory-exploration training paradigm called DyME, enabling small-scale vision-language models (SVLM) to acquire reasoning capabilities.
Enabling arbitrary inference in spatio-temporal dynamic systems: A physics-inspired perspective
Yan Ge, Yang Wang (University Of Science And Technology Of China)
Computational EfficiencyConvolutional Neural NetworkGraph Neural NetworkGraphTime SeriesPhysics Related
🎯 What it does: Proposed the PhySTA framework, combining Graph-Time Fourier Neural Operator with a multi-scale adaptive interaction module to model continuous spatiotemporal dynamics in graph structures and enable arbitrary region inference.
Enabling Fine-Tuning of Direct Feedback Alignment via Feedback-Weight Matching
Yunseok Lee (UNIST), Seulki Lee (KAIST)
ClassificationConvolutional Neural NetworkTransformerSupervised Fine-TuningImageText
🎯 What it does: Proposed the feedback-weight matching method, enabling direct feedback alignment (DFA) to reliably fine-tune pre-trained networks.
Enabling True Global Perception in State Space Models for Visual Tasks
Jie Hui (Xi'an Jiaotong University), Jianji Wang (Xi'an Jiaotong University)
Object DetectionSegmentationImage
🎯 What it does: Define global modeling theoretically, design GSSM and implement the GMamba module, using 2D-DFT pre-modulation to enhance the global perception of SSM, thereby achieving efficient and plug-and-play visual global modeling.
Enabling Your Forensic Detector Know How Well It Performs on Distorted Samples
Bin Li (Shenzhen University), bo.cao
Object DetectionSupervised Fine-TuningImage
🎯 What it does: Propose a distortion-aware confidence model (DACOM) for detecting image generation under the influence of noise, compression, and other distortions, combining it with traditional detectors to achieve sample-level confidence estimation.
ENACT: Evaluating Embodied Cognition with World Modeling of Egocentric Interaction
Qineng Wang (Northwestern University), Manling Li (Northwestern University)
Data SynthesisRobotic IntelligenceVision-Language-Action ModelWorld ModelTextBenchmark
🎯 What it does: Propose the ENACT benchmark, evaluating VLM's embodied cognition capabilities through forward and backward world modeling;
End-to-end Listen, Look, Speak and Act
Siyin Wang (Tsinghua University), Chao Zhang (ByteDance)
Robotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsVision-Language-Action ModelVideoTextMultimodalityAudio
🎯 What it does: Propose an end-to-end, full-duplex, streaming multimodal model ELLSA, capable of simultaneously listening, observing, speaking, and performing actions, supporting turn-taking in dialogues, immediate action interruption, and integrated question-answering combining speech and vision.
End-to-End Probabilistic Framework for Learning with Hard Constraints
Utkarsh Utkarsh (MIT CSAIL), Bernie Wang
OptimizationTime SeriesPhysics Related
🎯 What it does: Propose the ProbHardE2E framework, achieving end-to-end probabilistic prediction while strictly satisfying hard constraints and providing uncertainty quantification.
Endowing GPT-4 with a Humanoid Body: Building the Bridge Between Off-the-Shelf VLMs and the Physical World
Yingzhao Jian (Zhejiang University), Hehe Fan (Zhejiang University)
Robotic IntelligenceTransformerReinforcement LearningVision Language ModelDiffusion modelAuto EncoderMultimodality
🎯 What it does: Propose the BiBo framework, leveraging existing Vision-Language Models (e.g., GPT-4o) and diffusion models to control humanoid robots to perform diverse physical interaction tasks.
Energy-Based Transformers are Scalable Learners and Thinkers
Alexi Gladstone (University of Virginia), Tariq Iqbal (University of Virginia)
OptimizationComputational EfficiencyRepresentation LearningTransformerDiffusion modelScore-based ModelImageText
🎯 What it does: Propose Energy-Based Transformers (EBT), which verify input-prediction pairs by learning an energy function and implement System 2 thinking via gradient descent, enabling the model to automatically generate and self-verify predictions in unsupervised learning.
Energy-Efficient Random Variate Generation via Compressed Lookup Tables
Johann Ukrow (Hasso Plattner Institute University of Potsdam), Ralf Herbrich (Hasso Plattner Institute University of Potsdam)
GenerationCompressionComputational Efficiency
🎯 What it does: Proposes a fast and energy-efficient random variate generation method based on a compressed lookup table (cLUT), capable of sampling from arbitrary discrete distributions with controllable precision.
Energy-oriented Diffusion Bridge for Image Restoration with Foundational Diffusion Models
Jinhui HOU (City University of Hong Kong), Junhui Hou (City University of Hong Kong)
RestorationSupervised Fine-TuningDiffusion modelRectified FlowImage
🎯 What it does: Proposed the Energy-oriented Diffusion Bridge (E-Bridge) framework, achieving high-quality image restoration with one-time or minimal steps by constructing low-energy geodesic trajectories and a consistency solver;
Energy-Regularized Sequential Model Editing on Hyperspheres
Qingyuan Liu (Columbia University), Nanyun Peng (University Of Science And Technology Of China)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: In sequential editing of large-scale language models, the SPHERE stabilization strategy based on spherical energy (HE) is proposed and implemented, which can maintain the uniformity of weight distributions and reduce catastrophic forgetting during multiple knowledge updates;
Enforcing Axioms for AI Alignment under Loss-Based Rules
Alexandros Hollender (University of Oxford), Sonja Kraiczy (University of Oxford)
OptimizationData-Centric Learning
🎯 What it does: This paper investigates how principle-guided (constitutional-style) alignment methods may violate important social choice axioms (particularly Pareto optimality) within a reward learning framework based on loss, and explores scenarios across broader reward spaces (e.g., polynomials) and varying data distributions.
Enhanced Continual Learning of Vision-Language Models with Model Fusion
Haoyuan Gao (Shanghai Jiao Tong University), Weiran Huang (Tencent)
Meta LearningSupervised Fine-TuningMixture of ExpertsVision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: Introduces the ConDU framework, which uses model fusion to achieve continuous learning in vision-language models, compatible with full-parameter fine-tuning and parameter-efficient fine-tuning, and supports zero-shot inference.
Enhanced Generative Model Evaluation with Clipped Density and Coverage
Nicolas Salvy (Inria), Bertrand Thirion (Inria)
GenerationImage
🎯 What it does: Proposed and evaluated two new generative model quality assessment metrics: Clipped Density and Clipped Coverage.
Enhancing Communication Compression via Discrepancy-aware Calibration for Federated Learning
Zhiyi Wan (Beijing University Of Posts And Telecommunications), Xiaoqi Qin (Peng Cheng Laboratory)
Federated LearningComputational EfficiencyConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a difference-aware communication compression method based on locally calibrated data, significantly reducing communication volume and improving model accuracy in federated learning.
Enhancing Diffusion-Based Sampling with Molecular Collective Variables
Juno Nam (MIT), Benjamin Kurt Miller (FAIR at Meta)
Drug DiscoveryDiffusion modelScore-based ModelBiomedical DataPhysics Related
🎯 What it does: Proposed a Diffusion-based Sampler (WT-ASBS) that integrates Well-Tempered Bias in molecular sampling, achieving more efficient free energy curve reconstruction and reaction energy landscape sampling.
Enhancing Generative Auto-bidding with Offline Reward Evaluation and Policy Search
Zhiyu Mou (Taobao & Tmall Group of Alibaba), Bo Zheng (Taobao & Tmall Group of Alibaba)
Recommendation SystemTransformerReinforcement LearningSequential
🎯 What it does: Propose AIGB-Pearl, which integrates offline reinforcement learning with generative models, using a trajectory evaluator and KL-Lipschitz constraints to enable safe exploration and performance enhancement of generated trajectories.
Enhancing Geometric Perception in VLMs via Translator-Guided Reinforcement Learning
Hao Yu (Tsinghua University Guangdong Laboratory of Artificial Intelligence and Digital Economy), Chun Yuan (Tsinghua University Guangdong Laboratory of Artificial Intelligence and Digital Economy)
OptimizationSupervised Fine-TuningReinforcement LearningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Propose the GEOPERCEIVE benchmark and GEODSL language for precise evaluation of geometric perception in vision-language models; propose the GEODPO method, which uses an NL-DSL translator to generate reward signals and combines DPO for reinforcement learning to enhance geometric perception;
Enhancing Hallucination Detection through Noise Injection
Litian Liu (Qualcomm AI Research), Roland Memisevic (Qualcomm AI Research)
Anomaly DetectionComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Enhancing hallucination detection by injecting noise into the intermediate layers of LLMs during inference to capture model uncertainty.
Enhancing Image-Conditional Coverage in Segmentation: Adaptive Thresholding via Differentiable Miscoverage Loss
Rui Luo (City University of Hong Kong), Suqun Cao (City University of Hong Kong)
SegmentationConvolutional Neural NetworkSupervised Fine-TuningImageBiomedical Data
🎯 What it does: Proposes two adaptive threshold learning methods for image segmentation (AT and COAT), achieving image-level coverage control.
Enhancing Instruction Following of LLMs via Activation Steering with Dynamic Rejection
Minjae Kang (Yonsei University), Jaehyung Kim (Yonsei University)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Proposes a dynamic activation regulation method called DIRECTER to enhance large language models' (LLM) instruction-following capability during inference while avoiding text quality degradation caused by over-regulation.
Enhancing Language Model Reasoning with Structured Multi-Level Modeling
Siheng Xiong (Georgia Institute of Technology), Faramarz Fekri (Georgia Institute of Technology)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextChain-of-Thought
🎯 What it does: Proposes a Multi-Level Reasoning (MLR) framework that decomposes the reasoning process into high-level planning descriptors and low-level detailed reasoning, forming an alternating plan-execute cycle; simultaneously introduces an iterative Step-DPO training process and employs Twisted Sequential Monte Carlo (TSMC) to construct process-level preferences, addressing the long-sequence credit assignment challenge caused by sparse single-result rewards.
Enhancing Learning with Noisy Labels via Rockafellian Relaxation
Louis Chen, Johannes O. Royset
OptimizationAdversarial AttackHyperparameter SearchData-Centric LearningImage
🎯 What it does: Proposes the Rockafellian Relaxation Method (RRM), which enhances the robustness of neural networks in label-noisy environments by adaptively reweighting the training loss.
Enhancing LLMs for Knowledge Base Question Answering by Chain-of-Decomposition
Yonggang Zhang (Hong Kong University of Science and Technology), Jie Lu (University of Technology Sydney)
Computational EfficiencyTransformerTextGraphBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the Chain-of-Decomposition (CoD) framework, decomposing the KBQA task into three steps: retrieval, reconstruction, and reasoning, reducing the burden on LLMs and improving performance.
Enhancing Molecular Property Predictions by Learning from Bond Modelling and Interactions
Yunqing LIU, Wenqi Fan (Hong Kong Polytechnic University)
Drug DiscoveryGraph Neural NetworkTransformerGraph
🎯 What it does: Proposed the DeMol dual-graph multi-scale interaction framework, explicitly modeling atoms and bonds along with their interactions.
Enhancing Multi-Image Understanding through Delimiter Token Scaling
Minyoung Lee (Sogang University), Junsuk Choe (Sogang University)
Computational EfficiencyRepresentation LearningTransformerPrompt EngineeringImageTabularBenchmark
🎯 What it does: Propose scaling the hidden states of image separator tokens in multi-image inputs to suppress cross-image information leakage and enhance the model's multi-image understanding capability.
Enhancing Multivariate Time Series Forecasting with Global Temporal Retrieval
Fanpu Cao (Hong Kong University of Science and Technology), Hui Xiong (Hong Kong University of Science and Technology)
Convolutional Neural NetworkTime Series
🎯 What it does: Propose a lightweight, plug-and-play Global Temporal Retriever (GTR) module to expand the temporal awareness range of multivariate time series prediction models, thereby capturing long-term global periodic patterns.
Enhancing Persona Following at Decoding Time via Dynamic Importance Estimation for Role-Playing Agents
Yuxin Liu (University of Science and Technology of China), Lei Zhang (University of Science and Technology of China)
GenerationTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Propose the PDD framework, which utilizes conditional mutual information (CMI) for self-supervised estimation of context-related personality attribute importance, and during inference dynamically guides generation through weighted rewards, achieving role-playing agent personality following without fine-tuning.
Enhancing Sparse Event Detection in Healthcare Time-Series via Adaptive Gate of Context–Detail Interaction
Beomjun Bark (Seers Technology), Yun Kwan Kim (Seers Technology)
Anomaly DetectionTransformerReinforcement LearningTime SeriesBiomedical DataElectronic Health RecordsElectrocardiogram
🎯 What it does: Proposed a DETR-based time series multi-event detection framework that can simultaneously predict event types and their temporal boundaries.
Enhancing Stability of Physics-Informed Neural Network Training Through Saddle-Point Reformulation
Dmitry Bylinkin (Basic Research of Artificial Intelligence Laboratory), Aleksandr Beznosikov (Basic Research of Artificial Intelligence Laboratory)
OptimizationBenchmarkPhysics Related
🎯 What it does: Reformulate the training problem of physics-informed neural networks (PINNs) as a non-convex-strong convex saddle point problem and solve it using Bregman approximation;
Enhancing Trustworthiness of Fine-Tuned LLMs via Regularized Subset Selection
Kumar Shubham (Indian Institute of Science), Prathosh AP
Explainability and InterpretabilityData-Centric LearningReinforcement Learning from Human FeedbackLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Enhancing the reliability of LLMs by identifying and repairing training samples after SFT.
Enhancing Vision Transformers for Object Detection via Context-Aware Token Selection and Packing
Tianyi Zhang (University of Minnesota), Yu Cao (University of Minnesota)
Object DetectionTransformerImage
🎯 What it does: Proposed a sparse attention mechanism (SPA) based on learnable gating and packing, dynamically selecting and aggregating meaningful tokens in vision Transformers to enhance inference and training efficiency.
Enhancing Visual Token Representations for Video Large Language Models via Training-free Spatial-Temporal Pooling and Gridding
Bingjun Luo (Tsinghua University), Xinpeng Ding (Xidian University)
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelVision Language ModelVideo
🎯 What it does: Proposes ST-GridPool, a training-agnostic visual token enhancement method to improve visual representations in video large language models;
Enough is as good as a feast: A Comprehensive Analysis of How Reinforcement Learning Mitigates Task Conflicts in LLMs
Zixuan Ren (State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences), Chengqing Zong (Baidu Inc.)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: The study investigates merging models trained with reinforcement learning (RL) and traditional supervised fine-tuning (SFT) in large language models (LLMs), and systematically analyzes the impact of the two training paradigms on task conflict and post-merging performance.
Ensemble Prediction of Task Affinity for Efficient Multi-Task Learning
Afiya Ayman (Pennsylvania State University), Aron Laszka (Pennsylvania State University)
OptimizationComputational EfficiencyImageTabularTime Series
🎯 What it does: Propose the ETAP framework, which utilizes gradient similarity to compute task affinity, combines nonlinear mapping and residual correction to predict performance gains of task groups in multi-task learning, and efficiently selects task groups based on this to reduce negative transfer.
EnsembleSHAP: Faithful and Certifiably Robust Attribution for Random Subspace Method
Yanting Wang (Pennsylvania State University), Jinyuan Jia (Pennsylvania State University)
Explainability and InterpretabilityComputational EfficiencyAdversarial AttackImageText
🎯 What it does: Propose EnsembleSHAP, an efficient and provably secure feature attribution method specifically designed for random subspace methods.
Ensembling Pruned Attention Heads For Uncertainty-Aware Efficient Transformers
Firas Gabetni, Gianni Franchi (Institut Polytechnique De Paris)
ClassificationTransformerMixture of ExpertsImageText
🎯 What it does: Proposes Hydra Ensembles, an efficient ensemble method that achieves uncertainty estimation comparable to or better than Deep Ensembles by structurally pruning Transformer attention heads and fusing the pruned subnetworks into a single model, while maintaining inference speed and memory usage close to those of a single model.