Zero-Knowledge Proof Based Verifiable Inference of Models

2025-11-24

AI Summary

A fundamental tension exists between protecting AI model intellectual property and enabling verifiable inference. Existing zero-knowledge (ZK) verification approaches for neural networks support only limited operators, failing to cover full-layer semantics—including matrix multiplication, LayerNorm, Softmax, and SiLU—thus hindering end-to-end verifiable deep learning. Method: We propose the first ZK verification framework supporting complete neural network layers. It builds upon a trusted-setup-free, recursively composable zkSNARK architecture and employs the Fiat–Shamir heuristic to yield succinct non-interactive proofs. Our framework unifies circuit modeling for both linear and nonlinear layers and enables efficient proof generation. Contribution/Results: We instantiate our framework as ZK-DeepSeek—a fully verifiable variant of DeepSeek—demonstrating practical proof efficiency and flexibility under realistic workloads. This work establishes the first end-to-end, parameter-hiding verification protocol for deep learning inference, advancing both verifiable AI and model IP protection toward practical deployment.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Large Vision Models
• Natural Language Processing: (Large) Language Models

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Security and Privacy: Data transparency and provenance
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Recent advances in artificial intelligence (AI), particularly deep learning, have led to widespread adoption across various applications. Yet, a fundamental challenge persists: how can we verify the correctness of AI model inference when model owners cannot (or will not) reveal their parameters? These parameters represent enormous training costs and valuable intellectual property, making transparent verification difficult. In this paper, we introduce a zero-knowledge framework capable of verifying deep learning inference without exposing model internal parameters. Built on recursively composed zero-knowledge proofs and requiring no trusted setup, our framework supports both linear and nonlinear neural network layers, including matrix multiplication, normalization, softmax, and SiLU. Leveraging the Fiat-Shamir heuristic, we obtain a succinct non-interactive argument of knowledge (zkSNARK) with constant-size proofs. To demonstrate the practicality of our approach, we translate the DeepSeek model into a fully SNARK-verifiable version named ZK-DeepSeek and show experimentally that our framework delivers both efficiency and flexibility in real-world AI verification workloads.

Problem

• Verifying AI model inference without exposing proprietary parameters
• Enabling trust in deep learning outputs through zero-knowledge proofs
• Creating efficient verifiable neural networks including nonlinear operations

Innovation

• Zero-knowledge proofs verify AI inference correctness
• Framework supports linear and nonlinear neural network layers
• SNARK-verifiable DeepSeek model demonstrates practical efficiency