all AI news
High-dimensional limit theorems for SGD: Effective dynamics and critical scaling. (arXiv:2206.04030v1 [stat.ML])
June 9, 2022, 1:11 a.m. | Gerard Ben Arous, Reza Gheissari, Aukosh Jagannath
stat.ML updates on arXiv.org arxiv.org
We study the scaling limits of stochastic gradient descent (SGD) with
constant step-size in the high-dimensional regime. We prove limit theorems for
the trajectories of summary statistics (i.e., finite-dimensional functions) of
SGD as the dimension goes to infinity. Our approach allows one to choose the
summary statistics that are tracked, the initialization, and the step-size. It
yields both ballistic (ODE) and diffusive (SDE) limits, with the limit
depending dramatically on the former choices. Interestingly, we find a critical
scaling regime …
More from arxiv.org / stat.ML updates on arXiv.org
Jobs in AI, ML, Big Data
Senior Data Engineer
@ Publicis Groupe | New York City, United States
Associate Principal Robotics Engineer - Research.
@ Dyson | United Kingdom - Hullavington Office
Duales Studium mit vertiefter Praxis: Bachelor of Science Künstliche Intelligenz und Data Science (m/w/d)
@ Gerresheimer | Wackersdorf, Germany
AI/ML Engineer (TS/SCI) {S}
@ ARKA Group, LP | Aurora, Colorado, United States
Data Integration Engineer
@ Find.co | Sliema
Data Engineer
@ Q2 | Bengaluru, India