Investigating Blockchain as a Confirmation Mechanism for Decentralized Computing Platform’s PrivacyPreserving Deep Learning Models

Emergence of decentralized computing infrastructures with the advent of privacy-preserving deep learning is a paradigm shift in data-centric models training and deployment. But maintaining the integrity, trust, and transparency of the computations in these decentralized environments presents an open challenge. Benefits of Blockchain Technology for Digital Identity Onboarding Blockchain technology has the advantages of decentralized ledger and cryptographic verification methods which provide an effective means of overcoming these worries. This study explores using the blockchain as a verification mechanism for privacy-preserving deep learning models across distributed computing platforms. More specifically, we investigate in this work how the interaction with a blockchain network can be used to validate correctness of model updates, preserve data privacy, and generate trust in a non-centralized manner. We focus on ensuring the security of the training process through blockchain-based solutions while leveraging advanced techniques, like Federated Learning and Differential Privacy to preserve the privacy of sensitive data. Blockchain is used through a combination of smart contracts and consensus algorithms to record and verify every step during the model’s training cycle, with the goal of preventing malicious participants from tampering with or corrupting the model. The second part of the study investigates whether such an integration can scale by evaluating the efficiency of blockchain, given the large number of model update transactions during training. Experimental results show that a blockchain is a suitable layer with transparency and guaranteed correctness that can be used alone or in a multi-party setup that simultaneously increases the security and privacy of decentralized AI. This work paves the way for more secure decentralized AI systems by making sure that future privacypreserving deep learning models are both secure and scalable.