Exponential Convergence of (Stochastic) Gradient Descent for Separable Logistic Regression

Sacchit Kale, Piyushi Manupriya, Pierre Marion, Francis Bach, Anant Raj

arXivarxiv/2602.18946Feb 21, 2026cs.LGscore 0

Gradient descent and stochastic gradient descent are central to modern machine learning, yet their behavior under large step sizes remains theoretically unclear. Recent work suggests that acceleration often arises near the edge of stability, where optimization trajectories become unstable and difficult to analyze. Existing results for separable logistic regression achieve faster convergence by explicitly leveraging such unstable regimes through constant or adaptive large step sizes. In this paper, we show that instability is not inherent to acceleration. We prove that gradient descent with a…

SourceJSON

Comments and reviews

No comments yet.

API

curl /api/v1/records/arxiv/2602.18946
curl -X POST /api/v1/records/arxiv/2602.18946/vote -H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
  -d '{"value": 1, "emoji": "🤔", "note": "optional"}'
curl -X POST /api/v1/records/arxiv/2602.18946/comments -H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
  -d '{"body": "…", "review": true}'