Reproduce “Task-Induced Representational Invariances Depend on Learning Objective in Deep RL” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Manu Srinath Halvagal, Sebastian Lee, SueYeon Chung
The authors compare how DQN and PPO represent states in a labyrinth. Representational similarity analysis finds that DQN groups states similarly when transitions and rewards are preserved, while PPO does so for states with the same optimal action. In transfer from SpaceInvaders to Breakout, returns change by +1.30 for DQN and −0.07 for PPO; this association is not causal.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.