Pith. sign in

REVIEW 3 cited by

Convergence Rates for Stochastic Approximation: Biased Noise with Unbounded Variance, and Applications

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2312.02828 v4 pith:2K6OGPYY submitted 2023-12-05 stat.ML cs.LGmath.OCmath.PR

classification stat.MLcs.LGmath.OCmath.PR
keywords convergencecdotfunctionratestochasticassumptionsfunctionsglobal
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
abstract

In this paper, we study the convergence properties of the Stochastic Gradient Descent (SGD) method for finding a stationary point of a given objective function $J(\cdot)$. The objective function is not required to be convex. Rather, our results apply to a class of ``invex'' functions, which have the property that every stationary point is also a global minimizer. First, it is assumed that $J(\cdot)$ satisfies a property that is slightly weaker than the Kurdyka-Lojasiewicz (KL) condition, denoted here as (KL'). It is shown that the iterations $J(\boldsymbol{\theta}_t)$ converge almost surely to the global minimum of $J(\cdot)$. Next, the hypothesis on $J(\cdot)$ is strengthened from (KL') to the Polyak-Lojasiewicz (PL) condition. With this stronger hypothesis, we derive estimates on the rate of convergence of $J(\boldsymbol{\theta}_t)$ to its limit. Using these results, we show that for functions satisfying the PL property, the convergence rate of both the objective function and the norm of the gradient with SGD is the same as the best-possible rate for convex functions. While some results along these lines have been published in the past, our contributions contain two distinct improvements. First, the assumptions on the stochastic gradient are more general than elsewhere, and second, our convergence is almost sure, and not in expectation. We also study SGD when only function evaluations are permitted. In this setting, we determine the ``optimal'' increments or the size of the perturbations. Using the same set of ideas, we establish the global convergence of the Stochastic Approximation (SA) algorithm under more general assumptions on the measurement error, compared to the existing literature. We also derive bounds on the rate of convergence of the SA algorithm under appropriate assumptions.

Discussion (0). Sign in to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Stochastic Approximation in Banach Spaces Without Geometric Constraints

    math.PR 2026-07 accept novelty 6.5 of 10

    Almost-sure stochastic approximation holds on every Banach space for i.i.d. mean-zero noise (and for independent tight noise under moment conditions), with no geometric hypotheses required.

  2. Convergence Rate in Nonlinear Two-Time-Scale Stochastic Approximation with State (Time)-Dependence

    math.OC 2025-09 conditional novelty 6.0 of 10

    Under state- or time-dependent noise, two-time-scale stochastic approximation converges at rate O(k^{-t}) with t set by the noise decay exponents, and exponentially when state noise is exactly quadratic in the error.

  3. Convergence of Momentum-Based Optimization Algorithms with Time-Varying Parameters

    math.OC 2025-06 conditional novelty 6.0 of 10

    A unified momentum-based optimization algorithm with time-varying parameters is shown to converge almost surely under generalized Robbins-Monro and Kiefer-Wolfowitz-Blum conditions, even with biased, unbounded-varianc...

Pith tools