21 papers
LeAct: Learning to Reason from Expert Actions
Ziran Yang, Chengshuai Shi, Raj Ghugare +3
Modern reasoning models depend on reasoning data, today sourced from human annotations or distilled from stronger LLMs. However, a rich and largely untapped source of supervision l…
Can We Really Learn One Representation to Optimize All Rewards?
Chongyi Zheng, Royina Karegoudra Jayanth, Benjamin Eysenbach
As unsupervised pretraining becomes increasingly ubiquitous in reinforcement learning, a more thorough theoretical understanding of these methods becomes of equal importance to the…
Self-Supervised Goal-Reaching Results in Multi-Agent Cooperation and Exploration
Chirayu Nimonkar, Shlok Shah, Catherine Ji +1
For groups of autonomous agents to achieve a particular goal, they must engage in coordination and long-horizon reasoning. Rather than relying on complex reward functions and expli…
On the Role of Computation in Reinforcement Learning
Raj Ghugare, MichaÅ Bortkiewicz, Alicja Ziarko +1
How does the amount of compute available to a reinforcement learning (RL) policy affect its learning? Can policies using a fixed amount of parameters, still benefit from additional…
BuilderBench: The Building Blocks of Intelligent Agents
Raj Ghugare, Roger Creus Castanyer, Catherine Ji +4
Today's AI models learn primarily through mimicry and refining, so it is not surprising that they struggle to solve problems beyond the limits set by existing data. To solve novel…
Normalizing Flows are Capable Models for Continuous Control
Raj Ghugare, Benjamin Eysenbach
Modern reinforcement learning (RL) algorithms have found success by using powerful probabilistic models, such as transformers, energy-based models, and diffusion/flow-based models.…