ML2SC: Deploying Machine Learning Models as Smart Contracts on the Blockchain

Blockchain can increase AI transparency and trustworthiness: The paper demonstrates that machine learning models can be deployed on blockchain as "smart contracts," creating a permanent, tamper-proof record of how AI decisions are made. This addresses growing concerns about AI safety by making model computations verifiable and transparent to all parties.

Automatic translation bridges the AI-blockchain gap: ML2SC automatically converts machine learning models from Python (PyTorch) into Solidity smart contracts, eliminating the need for manual recoding. This technical bridge makes it practical for AI developers to deploy models on blockchain without requiring blockchain programming expertise.

On-chain AI comes with significant cost trade-offs: The paper shows that deploying and running ML models on blockchain incurs substantial "gas costs" (computational fees) that increase linearly with model complexity. This means larger, more capable AI models become prohibitively expensive to run on-chain, limiting which applications are practical.

Accuracy is preserved in the translation process: The paper confirms that models translated to smart contracts produce identical classification outputs to their original PyTorch versions, ensuring that the transparency gained through blockchain deployment doesn't compromise AI performance or reliability.

This approach is currently limited to simple neural networks: ML2SC works specifically with multi-layer perceptrons (MLPs), which are relatively simple models. More complex modern AI architectures would require additional development, limiting current real-world applicability.