Workshop
VerifAI @ ICLR
Layerwise Zero-Knowledge Proofs for Verifiable Large Language Model Inference
Zhaohui Wang
VerifAI Workshop at the International Conference on Learning Representations (ICLR), 2026
// ABSTRACT
The preliminary version of the NanoZK work, presented at the VerifAI workshop at ICLR 2026. It introduces layerwise zero-knowledge proofs that make outsourced LLM inference cryptographically verifiable: a prover commits to each transformer layer's computation and emits a succinct proof that a verifier checks in milliseconds, without learning the model weights or re-running inference.
The ICICS 2026 main-track paper significantly revises and extends this version with verifier-cost analysis, an IVC comparison, a lookup-table comparison, and GPU-projected scaling to GPT-2. Readers wanting the complete results should prefer that version; this entry records the workshop publication in its own right.
The ICICS 2026 main-track paper significantly revises and extends this version with verifier-cost analysis, an IVC comparison, a lookup-table comparison, and GPU-projected scaling to GPT-2. Readers wanting the complete results should prefer that version; this entry records the workshop publication in its own right.
// BIBTEX
@inproceedings{wang2026layerwise,
title = {Layerwise Zero-Knowledge Proofs for Verifiable Large Language Model Inference},
author = {Zhaohui Wang},
booktitle = {VerifAI Workshop at the International Conference on Learning Representations (ICLR)},
year = {2026},
month = {4},
eprint = {2603.18046},
archivePrefix = {arXiv},
}