Blockchain-Assisted Federated Learning

Centralized AI Problem

In the current Artificial Intelligence Era, the success of large-scale AI is dominated by a few big players, leading to a high degree of centralization. In such companies, precious datasets are mined through many facets, some of which may compromise intellectual property or other grey area datasets. Not only is this harmful for the future of AI legality, but this centralization tendency incentivizes monopoly from the government and other mega companies on their dataset repository, where a handful of gatekeepers could dictate the future of this transformative technology.

Blockchain Environment as Trustworthy Platform for Federated Learning

Researchers aim to mitigate this tendency by pushing the notion of Federated Learning, which promises a more open and collaborative effort by enabling training on vast, high-quality datasets that cannot be shared directly due to privacy, regulatory, or copyright restrictions.

To complement the previous technology, blockchain as a platform comes as a promising solution, ensuring a self-regulated environment for Decentralized AI in the future. Leveraging its decentralized property, it is possible to have an incentive mechanism that is self-sufficient, providing a novel marketplace for AI training in the future. However, many challenges need to be tackled first before achieving this goal, as efficiency, privacy, and security are not trivial to solve when we are using a permissionless blockchain such as Ethereum.

Our lab investigate multiple state-of-the-art approach to achieve this goal, such as succinct argument and incremental verifiable computation from applied cryptography field.