arXivIgor Ignashin, Anna Radovskaya, Andrew Semenov, Egor Lopatin, Stanislav Potapov, Aleksandr Kovalenko, Andrey Veprikov, Aleksandr Shestakov, Andrey Leonidov, Aleksandr BeznosikovThu, May 21, 2026, 8:50 AM PDT
score 14.7
SGD training is not random noise but structured landscape diffusion
Original: Why SGD is not Brownian Motion: A New Perspective on Stochastic Dynamics
Source: arxiv.org ↗
Writing ELI5 summary…