Reproduce “Hyperparameter and Neural Architecture Search via Evolutionary Optimization for Reinforcement Learning: HPNAS-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
Nihal Acharya Adde, Hanno Gottschalk
HPNAS extends architecture-only reinforcement-learning search by jointly evolving a policy’s visual-processing architecture and PPO training settings. In a Unity driving simulation, it balances reward, FLOPs, and parameter count. The fully trained joint-search policy reached a reported peak cumulative reward of 1205, versus 1190 for NAS-only and HPO-only policies and 1140 for the manually tuned baseline; seed variability was not reported.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.