Reproduce “Improved Training of Strong Stochastic Flow Maps” 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
Zander W. Blasingame, Antonio Franca, Raul Minan, Christian Koke, Niklas Rindtorff, Alexander Tong
iSSFM learns strong stochastic flow maps that jump between points along the same Brownian path. It approximates the rough path with shifted Legendre polynomials and uses a Wong–Zakai objective to enable derivative-based training, with a fine-tuning parameterization for a pre-trained flow model. On ImageNet-256, it records FD-DINOv2 143.0 at four neural-network evaluations, versus 475.1 for SSFM; this measures image quality, not wall-clock speed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.