Correctness-first PyTorch EBMs with FFT circulant layers, tiled Triton mixing, differentiable and Householder transforms, HRR formal-state search, Langevin sampling, and reproducible benchmarks.
benchmark deep-learning pytorch scientific-computing triton fourier formal-methods mcmc householder langevin-dynamics proof-search energy-based-model neuro-symbolic denoising-score-matching hrr structured-matrices circulant-matrix block-circulant differentiable-permutation spherical-langevin
-
Updated
Jul 14, 2026 - Python