ICLR 2026

5,370 papers

Top Research Directions

Research directions most represented by ICLR 2026 papers.

Scaling Language Model Reasoning Beyond Verifiable Rewards 88
Group Relative Policy Optimization for Large Reasoning Models 81
General Purpose Reasoning from Specialized Reinforcement Learning 79
Shifting from Static Benchmarks to Dynamic Evaluation 66
The Decoupling of Visual Perception and Logical Reasoning 51
Mitigating the Inference Latency of Chain of Thought 50
Activation Steering for Inference Time Model Control 41
Preventing Entropy Collapse in Reinforcement Learning 41
Inference Time Compute Scaling for Complex Reasoning 38
Reinforcement Learning with Verifiable Rewards 38
Autonomous Code Refinement via Execution Feedback Loops 38
Mitigating Computational Overthinking in Large Reasoning Models 37
Resolving Superposition with Sparse Autoencoders 34
Role Specialization in Multi-Agent Architectures 34
The Spatial Reasoning Bottleneck in Multimodal Models 33

Top Authors

AuthorPapers
Fei Huang20
Ge Zhang19
Jianye HAO19
Dahua Lin17
Junchi Yan17
Michael M. Bronstein17
Yejin Choi17
James Y. Zou16
James Zou16
Bo Han15
Jingren Zhou15
Min Zhang15
Tat-Seng Chua15
Caiming Xiong14
Dawn Song14
Pengfei Wan14
Yi Yang14
Jiaheng Liu13
Kai Chen13
Pengjun Xie13

Impactful Papers

Papers from ICLR 2026 that shaped research clusters.

Foundational

Papers

Title
Emergent Discrete Controller Modules for Symbolic Planning in Transformers
Emergent Hierarchical Reasoning in LLMs through Reinforcement Learning
Emergent Misalignment is Easy, Narrow Misalignment is Hard
EMFuse: Energy-based Model Fusion for Decision Making
EmoPrefer: Can Large Language Models Understand Human Emotion Preferences?
EmotionHallucer: Evaluating Emotion Hallucinations in Multimodal Large Language Models
Emotions Where Art Thou: Understanding and Characterizing the Emotional Latent Space of Large Language Models
EmotionThinker: Prosody-Aware Reinforcement Learning for Explainable Speech Emotion Reasoning
Empowering Efficiency and Efficacy in WebAgent via Enabling Info-Rich Seeking
Empowering LLM Tool Invocation with Tool-call Reward Model
Empowering Multi-Robot Cooperation via Sequential World Models
Empowering Small VLMs to Think with Dynamic Memorization and Exploration
Enabling arbitrary inference in spatio-temporal dynamic systems: A physics-inspired perspective
Enabling Fine-Tuning of Direct Feedback Alignment via Feedback-Weight Matching
Enabling True Global Perception in State Space Models for Visual Tasks
Enabling Your Forensic Detector Know ​How Well​ It Performs on Distorted Samples
ENACT: Evaluating Embodied Cognition with World Modeling of Egocentric Interaction
Endowing GPT-4 with a Humanoid Body: Building the Bridge Between Off-the-Shelf VLMs and the Physical World
End-to-end Listen, Look, Speak and Act
End-to-End Probabilistic Framework for Learning with Hard Constraints
Energy-Based Transformers are Scalable Learners and Thinkers
Energy-Efficient Random Variate Generation via Compressed Lookup Tables
Energy-oriented Diffusion Bridge for Image Restoration with Foundational Diffusion Models
Energy-Regularized Sequential Model Editing on Hyperspheres
Enforcing Axioms for AI Alignment under Loss-Based Rules
Enhanced Continual Learning of Vision-Language Models with Model Fusion
Enhanced Generative Model Evaluation with Clipped Density and Coverage
Enhancing Communication Compression via Discrepancy-aware Calibration for Federated Learning
Enhancing Diffusion-Based Sampling with Molecular Collective Variables
Enhancing Generative Auto-bidding with Offline Reward Evaluation and Policy Search
Enhancing Geometric Perception in VLMs via Translator-Guided Reinforcement Learning
Enhancing Hallucination Detection through Noise Injection
Enhancing Image-Conditional Coverage in Segmentation: Adaptive Thresholding via Differentiable Miscoverage Loss
Enhancing Instruction Following of LLMs via Activation Steering with Dynamic Rejection
Enhancing Language Model Reasoning with Structured Multi-Level Modeling
Enhancing Learning with Noisy Labels via Rockafellian Relaxation
Enhancing LLMs for Knowledge Base Question Answering by Chain-of-Decomposition
Enhancing Molecular Property Predictions by Learning from Bond Modelling and Interactions
Enhancing Multi-Image Understanding through Delimiter Token Scaling
Enhancing Multivariate Time Series Forecasting with Global Temporal Retrieval
Enhancing Persona Following at Decoding Time via Dynamic Importance Estimation for Role-Playing Agents
Enhancing Sparse Event Detection in Healthcare Time-Series via Adaptive Gate of Context–Detail Interaction
Enhancing Stability of Physics-Informed Neural Network Training Through Saddle-Point Reformulation
Enhancing Trustworthiness of Fine-Tuned LLMs via Regularized Subset Selection
Enhancing Vision Transformers for Object Detection via Context-Aware Token Selection and Packing
Enhancing Visual Token Representations for Video Large Language Models via Training-free Spatial-Temporal Pooling and Gridding
Enough is as good as a feast: A Comprehensive Analysis of How Reinforcement Learning Mitigates Task Conflicts in LLMs
Ensemble Prediction of Task Affinity for Efficient Multi-Task Learning
EnsembleSHAP: Faithful and Certifiably Robust Attribution for Random Subspace Method
Ensembling Pruned Attention Heads For Uncertainty-Aware Efficient Transformers

« prev Page 30/108 next »