Aviral Kumar

floq: Training Critics via Flow-Matching for Scaling Compute in Value-Based RL

RLAD: Training LLMs to Discover Abstractions for Solving Reasoning Problems

e3: Learning to Explore Enables Extrapolation of Test-Time Compute for LLMs

TRIM: Hybrid Inference via Targeted Stepwise Routing in Multi-Step Reasoning Tasks

InT: Self-Proposed Interventions Enable Credit Assignment in LLM Reasoning

Teaching LLMs When to Stop Seeking and Start Acting

Learning to Reason on Hard Problems with Privileged On-Policy Exploration

RLAC: Reinforcement Learning with Adversarial Critic for Free-Form Generation Tasks

Horizon Reduction Makes RL Scalable

Training Language Models to Self-Correct via Reinforcement Learning

Grounded Reinforcement Learning for Visual Reasoning

Scaling LLM Test-Time Compute Optimally Can be More Effective than Scaling Parameters for Reasoning

Bigger, Regularized, Categorical: High-Capacity Value Functions are Efficient Multi-Task Learners

Compute-Optimal Scaling for Value-Based Deep RL

Rewarding Progress: Scaling Automated Process Verifiers for LLM Reasoning

Optimizing Test-Time Compute via Meta Reinforcement Finetuning

What Do Learning Dynamics Reveal About Generalization in LLM Mathematical Reasoning?

Efficient Online Reinforcement Learning Fine-Tuning Need Not Retain Offline Data

RRM: Robust Reward Model Training Mitigates Reward Hacking

Reasoning as an Adaptive Defense for Safety

Thinking vs. Doing: Improving Agent Reasoning by Scaling Test-Time Interaction

Scaling Test-Time Compute Without Verification or RL is Suboptimal

Inference-Aware Fine-Tuning for Best-of-N Sampling in Large Language Models

Value-Based Deep RL Scales Predictably

Generative Verifiers: Reward Modeling as Next-Token Prediction

Digi-Q: Learning VLM Q-Value Functions for Training Device-Control Agents

Vision-Language Models Provide Promptable Representations for Reinforcement Learning

Improving the Efficiency of Test-Time Search in LLMs with Backtracking

Parameterization Agnostic RL

Pre-Memorization Train Accuracy Reliably Predicts Generalization in LLM Reasoning