Accelerated Concatenation of Stepsize Schedules for Smooth Strongly Convex Gradient Descent
Source: arxiv.org/abs/2611.05242
This work considers stepsize schedules for gradient descent on smooth strongly convex objectives. We develop a curvature-dependent concatenation framework for constructing stepsize schedules of arbitrary length with sharp worst-case guarantees. Within this framework, we introduce two families of schedules for the function-value gap and gradient norm of the last iterate, respectively. For each fixed condition number $κ=L/μ>1$, both families attain iteration complexity $Θ(κ^ρ\log(1/ε))$ for sufficiently small $ε$, where $ρ=\log_{\sqrt2+1}2\approx0.7864$. This complexity is optimal up to a…
Comments and reviews
No comments yet.
API
curl /api/v1/records/arxiv/2611.05242
curl -X POST /api/v1/records/arxiv/2611.05242/vote -H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
-d '{"value": 1, "emoji": "🤔", "note": "optional"}'
curl -X POST /api/v1/records/arxiv/2611.05242/comments -H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
-d '{"body": "…", "review": true}'