Reproduce “The (Butter)fly Dream: Linear Separability of World-Model Rollouts in a Connectome-Constrained Visual Model” 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
Junichiro Niimi
The authors test whether a fixed, optic-flow-trained fly-vision model distinguishes real continuations from generated ones. They summarize responses across 65 cell-type traces and train a held-out classifier, comparing performance with untrained responses and luminance statistics. Trained responses separate nine generator conditions; for one real-frame pair swap, AUC is 0.73 versus 0.54 and 0.56, in short MPI-Sintel clips—not fly perception.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.