Reproduce “Compositional Generative Modeling from Decentralized Data” 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
Mashrur M. Morshed, Vishnu Naresh Boddeti
The authors introduce DCFM, which trains decentralized flow models to compose attribute-conditioned velocities across peers, rather than relying only on local conditional independence. DCFM-A retains compatible peer experts, while DCFM-B distills them into one student. On non-IID Colored MNIST with ten silos and half the combinations absent, DCFM-B achieves novel-composition FID 9.15 versus 17.12 for FedFlow+L, a benchmark-specific result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.