Reproduce “Fractional Diffusion Bridge Models” 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

Gabriel Nobis, Maximilian Springenberg, Arina Belova, Rembert Daems, Christoph Knochenhauer, Manfred Opper, Tolga Birdal, Wojciech Samek

Fractional Diffusion Bridge Models (FDBM) extend stochastic bridges with approximations to fractional Brownian motion, using correlated increments whose roughness is controlled by the Hurst index. On the D3PM protein test set, FDBM had a median Cα RMSD of 2.12 Å and 48% below 2 Å, versus 2.40 Å and 43% for ABM; the comparison does not isolate memory as the cause.

Key insights

  • Approximating fractional Brownian motion with a higher-dimensional, Markovian augmented process (MA-fBM) resolves the intractability issue while retaining the non-Markovian characteristics of fBM.
  • The Hurst index (H) provides explicit control over the diffusion's roughness and memory effects, allowing the model to adapt its dynamics to match the intrinsic characteristics of different datasets.
  • Incorporating fractional noise into diffusion bridge models enables a more physically realistic representation of complex system dynamics, leading to improved generative fidelity.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.