TIGER: Tunable-Geometry Accelerated Gradient Sliding with Adaptive Safeguards under Similarity

Darina M. Dvinskikh, Anton G. Novitskii, Ewsey R. Obzherin, Alexander V. Gasnikov

arXivarxiv/2610.25504Oct 18, 2026math.OCscore 0

We study distributed composite optimization under function similarity when synchronized gradient queries dominate local computation. We propose TIGER, an accelerated gradient-sliding method in problem-adapted Bregman geometry. The method applies to constrained non-Euclidean domains, allows the communicated correction term to be nonconvex, and terminates each local subproblem using a computable stationarity residual. Each iteration reuses one synchronized correction gradient during the local solve and makes a second query to construct the composite subgradient for the mirror step. Under…

SourceJSON

Comments and reviews

No comments yet.

API

curl /api/v1/records/arxiv/2610.25504
curl -X POST /api/v1/records/arxiv/2610.25504/vote -H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
  -d '{"value": 1, "emoji": "🤔", "note": "optional"}'
curl -X POST /api/v1/records/arxiv/2610.25504/comments -H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
  -d '{"body": "…", "review": true}'