-
Expected Return Causes Outcome-Level Mode Collapse
Why reinforcement learning keeps giving the same answer even when many answers are good, and a one-line fix called inverse probability scaling.
-
MIPS: Mode Collapse in Large Outcome Spaces
Keeping RL diverse when there are far too many possible outcomes to count, by estimating each outcome's probability from a single sample.
-
Density-Aware Reward Scaling
Diverse images and text from standard RL, by giving less reward to outputs that look like many others in the same batch.
-
Matryoshka Transcoders
Automatically finding, and describing in plain English, the ways image generators break physics, such as extra fingers, warped text and impossible poses.
-
Editing Discrete Latent Variables with Reinforcement Learning
An RL agent that makes minimal, interpretable edits to discrete VQ-VAE codes to reach a target class.
-
Brain-Inspired Attention for Object Counting
A model that counts objects the way people do. It looks around in quick jumps, keeps a running tally, and stops once it is sure.
Projects
Research projects in reinforcement learning, representation learning and generative AI.