OmniLytics: A Blockchain-based Secure Data Market for Decentralized Machine Learning

Abstract

We propose OmniLytics, a blockchain-based secure data trading marketplace for machine learning applications. Utilizing OmniLytics, many distributed data owners can contribute their private data to collectively train a ML model requested by some model owners, and get compensated for data contribution. OmniLytics enables such model training while simultaneously providing 1) model security against curious data owners; 2) data security against curious model and data owners; 3) resilience to malicious data owners who provide faulty results to poison model training; and 4) resilience to malicious model owner who intents to evade the payment. OmniLytics is implemented as a smart contract on the Ethereum blockchain to guarantee the atomicity of payment. In Omni-Lytics, a model owner publishes encrypted initial model on the contract, over which the participating data owners compute gradients using their private data, and securely aggregate the gradients through the contract. Finally, the contract reimburses the data owners, and the model owner decrypts the aggregated model update. We implement a working prototype of OmniLytics on Ethereum, and perform extensive experiments to measure its gas cost and execution time under various parameter combinations, demonstrating its high computation and cost efficiency and strong practicality.

Lightweight Blockchain-Empowered Secure and Efficient Federated Edge Learning

“…The peer-to-peer distributed nature of blockchain makes the decentralized aggregation of global model available, shedding new light on handling the single point of failure in conventional FL. Research works [7]- [10], [12], [14], [17]- [21] move the aggregation step from the server to the blockchain nodes, while Qu et al [22] and Mugunthan et al [15] eliminate the role of the aggregator by letting clients themselves aggregate model updates obtained from the blockchain network. To ensure the reliability of the global model, most blockchained FL frameworks embed local model update verification into their design.…”

“…Instead of evaluating model updates with the help of clients, which would inevitably increase the communication burden and be restricted with clients' active status, recent blockchain-based FL frameworks reaped recent advances in Robust FL to protect against malicious clients. Biscotti [7], SPDL [11] and Omnilytics [14] applied the Multi-krum as the validation mechanism and give passes to model updates with closer Euclidean distances. Biscotti introduces noiser, verifier and aggregator nodes, where noisers produce differentially private (DP) Gaussian noise to be added to the model updates, verifiers run Multi-krum to sign commitments for passed updates, and aggregators aggregate unmasked passed model updates via a secure protocol.…”

“…The deployments of proposed blockchain-based FL systems in the research community roughly fall into three categories: to client devices [7]- [10], to edge nodes [12], [13], [23], and to the existing blockchain platform Ethereum with smart contracts [14], [15]. Considering the limited resources of client devices, running and maintaining blockchain locally is costly in terms of communication and storage for the large propagated block in the network and the ever-growing blockchain.…”

“…However, the ChainsFL [12] requires the sharing of a test dataset for local model update validation, which is not compatible with the privacypreserving objective of FL, while [13] is vulnerable to the single point of failure since the centralised aggregation is carried out only by the task publisher. Another branch of works uses a popular blockchain platform like Ethereum with smart contracts [14] [15], however, they are costly to operate, as every interaction with the blockchain consumes a monetary gas fee. Thus, there is a lack of a cost-effective, secure and reliable blockchain-based FL system that runs at vulnerable network edge with resource-constrained client devices.…”