Fokker-Planck Analysis and Invariant Laws for a Continuous-Time Stochastic Model of Adam-Type Dynamics
Kaj Nystr\"om

TL;DR
This paper models Adam-type optimization algorithms using continuous-time stochastic differential equations, analyzes their invariant measures, and proves exponential convergence under certain conditions.
Contribution
It introduces a continuous-time stochastic model for Adam-type methods, analyzes the associated Fokker-Planck equation, and establishes convergence properties.
Findings
Existence and uniqueness of invariant measures under mild conditions.
Explicit characterization of noise propagation via matrix A(x).
Proof of exponential convergence to invariant measure.
Abstract
We develop a continuous-time model for the long-term dynamics of adaptive stochastic optimization, focusing on bias-corrected Adam-type methods. Starting from a finite-sum setting, we identify a canonical scaling of learning rates, decay parameters, and gradient noise that yields a coupled, time-inhomogeneous stochastic differential equation for the parameters , first-moment tracker , and second-moment tracker . Bias correction persists via explicit time-dependent coefficients, and the dynamics becomes asymptotically time-homogeneous. We analyze the associated Fokker-Planck equation and, under mild regularity and dissipativity assumptions on , prove existence and uniqueness of invariant measures. Noise propagation is governed by . Hypoellipticity may fail on , where \[ \mathcal…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
