Central Limit Theorem for Two-Timescale Stochastic Approximation with Markovian Noise: Theory and Applications
17 January 2024
Do Young Eun
Establishes central limit theorem for nonlinear two-timescale stochastic algorithms
Uncovers effect of underlying Markov chain on asymptotic behavior
Expands sampling efficiency strategies from SGD to bilevel, minimax optimization
Shows nonlinear GTD algorithms have identical asymptotic covariance
Central limit theorem for two-timescale stochastic algorithms
This paper establishes a central limit theorem for two-timescale stochastic algorithms under Markovian noise. Key contributions include uncovering the coupled dynamics influenced by the Markov chain, expanding efficient sampling strategies from SGD to broader applications, and deducing identical asymptotic performance of nonlinear GTD algorithms.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper