[Paper Review] Trustless Machine Learning Contracts; Evaluating and Exchanging Machine Learning Models on the Ethereum Blockchain

A. Besir Kurtulmus, Kenny Daniel|arXiv (Cornell University)|Feb 27, 2018

Blockchain Technology Applications and SecurityComputer Science8 references107 citations

TL;DR

The paper proposes the DanKu protocol, a trustless Ethereum-based system to solicit, train, evaluate, and payout for ML models using hashed datasets and on-chain evaluation. It discusses implementation, incentives, threat models, and practical considerations for deploying ML contracts on Ethereum.

ABSTRACT

Using blockchain technology, it is possible to create contracts that offer a reward in exchange for a trained machine learning model for a particular data set. This would allow users to train machine learning models for a reward in a trustless manner. The smart contract will use the blockchain to automatically validate the solution, so there would be no debate about whether the solution was correct or not. Users who submit the solutions won't have counterparty risk that they won't get paid for their work. Contracts can be created easily by anyone with a dataset, even programmatically by software agents. This creates a market where parties who are good at solving machine learning problems can directly monetize their skillset, and where any organization or software agent that has a problem to solve with AI can solicit solutions from all over the world. This will incentivize the creation of better machine learning models, and make AI more accessible to companies and software agents.

Motivation & Objective

• Introduce a trustless marketplace protocol for exchanging machine learning models on the Ethereum blockchain.
• Define the DanKu (Dan iel + Ku rtulmus) contract protocol and its five-stage process.
• Show how cryptographic hashing, dataset randomization, and on-chain evaluation enable fair payouts.
• Discuss incentives, threat models, and practical considerations for implementing ML contracts on Ethereum.

Proposed method

• Define DanKu Contract (DKC) as an Ethereum contract implementing a five-stage lifecycle: initialization, submission, test data reveal, evaluation, and finalization.
• Use hashed data groups with nonces and sha3-keccak hashing to secure data integrity and prevent tampering.
• Randomly partition data groups into training and testing sets using previous block hashes as seeds.
• Provide a forward_pass implementation for a simple neural network to demonstrate on-chain evaluation.
• Implement an evaluation function that uses a chosen metric (e.g., accuracy) to score models submitted by participants.
• Address gas constraints and network considerations by allowing per-model evaluation and local/off-chain evaluation options.

Experimental results

Research questions

• RQ1 How can a trustless, decentralized protocol securely solicit, train, and evaluate ML models on the Ethereum blockchain?
• RQ2 What mechanisms ensure fair payout and prevent cheating by organizers or participants in ML model competitions?
• RQ3 What are the practical limitations (gas costs, data storage, math operations) of executing ML models on EVM, and how can they be mitigated?

Key findings

• The DanKu protocol enables anonymous, trustless ML model competitions with automatic payout based on on-chain evaluation results.
• Hashing datasets with nonces and using block hashes for data partitioning reduces organizer manipulation and rainbow-table attacks.
• Submitting and evaluating models can be performed in a way that mitigates too-many-submissions and gas-limit risks by per-model evaluation.
• There are significant practical considerations including data storage costs, lack of floating-point arithmetic, and model execution time within gas limits.
• The protocol discusses potential extensions such as homomorphic encryption, IPFS/swarm for data storage, and GPU miner arbitrage implications.