Reproduce “Bridging Neural Tangent Kernel Analysis and Reinforcement Learning for Deep Function Approximation” 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
William St-Arnaud, Margarida Carvalho, Golnoosh Farnadi
The authors extend neural tangent kernel analysis to temporal-difference learning with sequential mini-batches and refreshed target networks, deriving a closed-form solution in an infinite-width model. They use predicted values to guide solver searches over state-action pairs, adding candidates with high temporal-difference error. On a deterministic 8×8 grid, SQN with the solver scores 1.02 versus 0.74 at 1,000 steps; this does not establish gains beyond that setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.