Papers · Proving training · Optimum Vicinity

Optimum Vicinity

Founding Zero-Knowledge Proofs of Training on Optimum Vicinity

Gefei Tan, Adrià Gascón, Sarah Meiklejohn, Mariana Raykova, Xiao Wang, Ning Luo

Venue
ACM CCS 2025
Date
2025-01-13
Proof system
Bounds distance between the trained model and the true optimum (convex problems)
Numbers from
primary
Paper
https://eprint.iacr.org/2025/053
PDF
https://eprint.iacr.org/2025/053.pdf

Notes

A genuinely different paradigm, worth highlighting: instead of proving every training step was executed correctly, prove the result lies within a bounded vicinity of the mathematical optimum. Only applies where training is a convex optimization problem. Boolean circuits up to 246x smaller, arithmetic circuits up to 5x smaller than step-by-step zkPoT.

Our reading

Citation neighbourhood

Builds on
  • none recorded
Cited by, in this corpus
  • none recorded

Edges are a proxy: paper A's text mentions B anywhere (body or bibliography). See the full graph.

Discussed in

The hybrid cellWhat each system actually provesThe hardest claim in the tableEvery relaxation, and what an adversary can still doThe systems, grouped by what they gave up