DeepProve is a framework for proving neural network inference using zero-knowledge cryptography, targeting developers and researchers in AI and blockchain. It enables fast, verifiable computation of ML models like MLPs and CNNs without exposing sensitive data, enhancing privacy and trust in applications from healthcare to decentralized finance.

How It Works

DeepProve utilizes advanced cryptographic techniques, specifically sumchecks and logup GKR, to achieve sublinear proving times. This approach allows for significantly faster generation of proofs for model inferences compared to traditional methods, making zero-knowledge machine learning more practical and efficient.

Quick Start & Requirements

• Install: cargo build --release (from source)
• Prerequisites: Rust toolchain, C++ compiler.
• Resources: Requires significant CPU and RAM for proving.
• Docs: ZKML README

Highlighted Details

• Claims 158x faster proving for a 264k parameter CNN and 54x faster for a 4M parameter dense model compared to EZKL.
• Reuses sumcheck and GKR implementations from scroll-tech/ceno.
• Supports MLPs and CNNs.

Maintenance & Community

• Built upon work from scroll-tech/ceno.
• No explicit community links (Discord/Slack) or roadmap provided in the README.

Licensing & Compatibility

• zkml subcrate is under the "Lagrange License".
• The rest of the code is licensed under Apache 2.0 + MIT.
• The "Lagrange License" may have restrictions for commercial use or linking with closed-source projects; clarification is recommended.

Limitations & Caveats

The project's core proving logic is under a custom "Lagrange License," which may impose restrictions on commercial use or integration into proprietary systems. The README does not detail the specifics of this license.