Papers · Proving training · ZKBoost

ZKBoost

ZKBoost: Zero-Knowledge Verifiable Training for XGBoost

Nikolas Melissaris, Antigoni Polychroniadou, Akira Takahashi, Chenkai Weng, Jiayi Xu

Venue
IACR ePrint 2026/202
Date
2026-02-08
Affiliation
CNRS/IRIF & Université Paris Cité / J.P. Morgan AI Research / Arizona State University
Proof system
Generic zkPoT template + VOLE-based instantiation
Numbers from
primary
Paper
https://eprint.iacr.org/2026/202

Notes

First zkPoT for XGBoost -- a reminder that verifiable training is not only about deep nets, and gradient-boosted trees are what most tabular/finance production models are. Fixes a security gap in prior ZK training proofs. Fixed-point XGBoost variant matches standard XGBoost accuracy within 1%. Runtime table not yet extracted.

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 hardest claim in the tableEvery relaxation, and what an adversary can still doThe systems, grouped by what they gave up
Other recorded fields
arxiv: https://arxiv.org/abs/2602.04113