{"id":"arxiv/2610.25304","url":"https://arxiv.org/abs/2610.25304","tags":["math.OC"],"score":0,"title":"Variance Reduction Note 2: The Convergence of SPIDER","domain":"math.OC","source":"arxiv","authors":["Alina Lodhi"],"reviews":0,"subject":"math.OC","summary":"The stochastic path-integrated differential estimator (SPIDER) tracks a gradient by accumulating small, shared-sample gradient differences and periodically refreshing the estimate. This second expository note on recursive variance reduction explains SPIDER-SFO and gives a self-contained proof of its expected first-order stationarity guarantee. The argument combines an epochwise mean-square error bound with descent under a clipped, normalized update. It yields $O(ε^{-3})$ stochastic gradient evaluations in the online setting and $O(n+\\sqrt n\\,ε^{-2})$ component gradient evaluations for a…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-10-18","reactions":{}}