Aaron Courville

The Markovian Thinker: Architecture-Agnostic Linear Scaling of Reasoning

Generative Adversarial Post-Training Mitigates Reward Hacking in Live Human-AI Music Interaction

The Intricate Dance of Prompt Complexity, Quality, Diversity and Consistency in T2I Models

Simplicial Embeddings Improve Sample Efficiency in Actor–Critic Agents

Towards Sustainable Investment Policies Informed by Opponent Shaping

A Comedy of Estimators: On KL Regularization in RL Training of LLMs

Asymmetric Proximal Policy Optimization: mini-critics boost LLM reasoning

Shape of Thought: When Distribution Can Matter More than Correctness in Reasoning Tasks

Learning Robust Social Strategies with Large Language Models

Stable Gradients for Stable Learning at Scale in Deep Reinforcement Learning

Advantage Alignment Algorithms

Don't flatten, tokenize! Unlocking the key to SoftMoE's efficacy in deep RL

Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation

Measure gradients, not activations! Enhancing neuronal activity in deep reinforcement learning

Compositional Discrete Latent Code for High Fidelity, Productive Diffusion Models

Forgetting Transformer: Softmax Attention with a Forget Gate

BiXSE: Improving Dense Retrieval via Probabilistic Graded Relevance Distillation

The Courage to Stop: Overcoming Sunk Cost Fallacy in Deep Reinforcement Learning

Mitigating Plasticity Loss in Continual Reinforcement Learning by Reducing Churn

VinePPO: Refining Credit Assignment in RL Training of LLMs

Neuroplastic Expansion in Deep Reinforcement Learning

Scaling Stick-Breaking Attention: An Efficient Implementation and In-depth Study

Adaptive Computation Pruning for the Forgetting Transformer

The Impact of On-Policy Parallelized Data Collection on Deep Reinforcement Learning Networks

Sample, Predict, then Proceed: Self-Verification Sampling for Tool Use of LLMs

Asynchronous RLHF: Faster and More Efficient Off-Policy RL for Language Models

Not All LLM Reasoners Are Created Equal

FLAM: Frame-Wise Language-Audio Modeling

VinePPO: Unlocking RL Potential For LLM Reasoning Through Refined Credit Assignment

Bias Analysis in Unconditional Image Generative Models

Training Universal Text Encoders with Pair Relevance Classification Loss