Reproduce “A Bregman Proximal Viewpoint on Neural Operators” 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
Abdel-Rahim Mezidi, Jordan Patracone, Saverio Salzo, Amaury Habrard, Massimiliano Pontil, Remi Emonet, Marc Sebban
The authors introduce the Bregman Fourier neural operator (BFNO), deriving operator layers from regularized optimization over functions and adding an inverse-activation term before learned updates. On 2D Navier–Stokes at viscosity 10⁻⁴, a 16-layer BFNO with SoftPlus reports 12.2 ± 0.1% ℓ2 test relative error, versus 12.6 ± 0.1% for FNO with ReLU. The approximation theorem assumes invertible sigmoidal activations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.