Papers · Proving training · Kaizen

Kaizen

Zero-Knowledge Proofs of Training for Deep Neural Networks

Kasra Abbaszadeh, Christodoulos Pappas, Jonathan Katz, Dimitrios Papadopoulosunverified

Venue
ACM CCS 2024
Date
2024-02-02
Affiliation
University of Maryland / HKUST
Proof system
Recursive composition of GKR-style proofs + aggregatable polynomial commitments
Hardware
8x Xeon Platinum 8370, 512 GB RAM
Numbers from
primary
Paper
https://eprint.iacr.org/2024/162
PDF
https://eprint.iacr.org/2024/162.pdf

Reported benchmarks

modelparamsbatch sizeproving time sproof size mbverification time s
VGG-1110M169001.630.13

proving_time_note: ~15 min per training iteration (survey Table IV: <882s)

Notes

The reference zkPoT for DNNs. Prover trains via mini-batch gradient descent and emits a commitment plus succinct proof each iteration; iteration count need not be fixed in advance. Proof size and verifier time are independent of iteration count and dataset size. 24x faster than generic recursive proofs.

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