网安资讯详情 - SecLens 情报雷达

网安资讯,一网打尽。汇集权威漏洞通告与行业要闻,结合分组浏览、智能过滤、RSS订阅 和 Webhook 推送,多通道拓展您的安全情报视野。

Agree on the Model, Verify the Inference: GKR Protocols for HND-Based Transformer Inference

来源: arxiv_cs_cr · 发布时间 2026-07-23 18:52 (UTC+08:00) · 抓取时间 2026-07-26 19:10 (UTC+08:00)

原文链接

摘要

Outsourced Transformer inference exposes clients to model substitution and incomplete execution, while direct replay removes the computational benefit of delegation. We present GKR-HND, a registered-model protocol for verifying the polynomial backbone of Homomorphic--Nonhomomorphic Decomposition Transformers. The retained verifier checks the GKR transcript and registered-weight openings, but delegates expensive public evaluations to an assigned computation worker. Assuming an honest retained verifier and prover--worker non-collusion, the verifier accepts only when the worker's signed, request-bound response agrees with the proof claims. Experiments with pretrained HND models validate the proof path and the delegated public computation without dense-matrix replay.

正文

Outsourced Transformer inference exposes clients to model substitution and incomplete execution, while direct replay removes the computational benefit of delegation. We present GKR-HND, a registered-model protocol for verifying the polynomial backbone of Homomorphic--Nonhomomorphic Decomposition Transformers. The retained verifier checks the GKR transcript and registered-weight openings, but delegates expensive public evaluations to an assigned computation worker. Assuming an honest retained verifier and prover--worker non-collusion, the verifier accepts only when the worker's signed, request-bound response agrees with the proof claims. Experiments with pretrained HND models validate the proof path and the delegated public computation without dense-matrix replay. Authors: Xiaolong Liang, Juanjuan Li, Rui Qin, Yisheng Lv Categories: cs.LG, cs.CR PDF: https://arxiv.org/pdf/2607.21162v1 Comment: 24 pages, including 4 pages of supporting information; 2 figures

标签

扩展字段

{
  "arxiv_id": "2607.21162v1",
  "authors": [
    "Xiaolong Liang",
    "Juanjuan Li",
    "Rui Qin",
    "Yisheng Lv"
  ],
  "categories": [
    "cs.LG",
    "cs.CR"
  ],
  "comment": "24 pages, including 4 pages of supporting information; 2 figures",
  "doi": null,
  "entry_id": "https://arxiv.org/abs/2607.21162v1",
  "pdf_url": "https://arxiv.org/pdf/2607.21162v1",
  "primary_category": "cs.LG",
  "search_query": "cat:cs.CR",
  "updated_at": "2026-07-23T10:52:53+00:00"
}