Reproduce “PONDER: Learning to Generate in the Right Order” 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
Julius Erbach, Dominik Narnhofer, Bernt Schiele, Jan Eric Lenssen, Konrad Schindler
PONDER trains a policy to choose which image patches a frozen masked generator fills next and which earlier predictions it revisits. The policy uses the partly generated image and receives rewards based on completed samples. On hard visual MNIST Sudoku, exact-solution accuracy is 56.6% without rethinking and 96.7% with nine removals; this result is specific to the task and setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.