Get Started
Research questionWhat convergence guarantees and irreducible errors arise in random-reshuffling optimization for finite sums and composite objectives?Random reshuffling makes component updates dependent within each epoch, so convergence depends on how work is distributed across epochs. Composite objectives can also introduce a residual from the proximal splitting step.
Machine Learning
Research Paper
Statistical Machine Learning
Latest papersRecent research connected to this question, newest first.Centered Permutation Prefixes for SGD with Random Reshuffling: Sharp Rates, Hölder Geometry, and Composite Proximal ExtensionsThe source gives theoretical results under Lipschitz component gradients and a strongly convex average with Lipschitz- or Hölder-continuous Hessian; the components may be nonconvex in the main setting. It also treats convex-component decreasing stepsizes and epoch-wise proximal updates, with lower-bound and residual claims limited to the stated constant-stepsize regimes; the evidence is theoretical rather than deployment-based.research paper · Sep 4, 2026
Related questions
How can Adam’s error be bounded for strongly convex stochastic optimization without assuming bounded iterates?How can iterated one-step approximations control error in strongly continuous convex monotone semigroups?How can iterative data-consistent inversion recover joint dependence in generalized stochastic inverse problems?Can one squared-loss estimator achieve both minimax and universal exponential rates for finite versus countably infinite hypothesis classes?
Home
Topics
Search
Library