AI (ML) can be used in various fields. However, while data is essential for learning, a lot of data is private. AI companies collect and utilize such personal data, but accordingly, privacy infringement on people's data is severe, and proper compensation is not made. Solai offers innovative solutions to this based on Federated Learning.
Anybody can propose any task that one wants to train. If the proposal is passed, training starts. Using the Federated Learning scheme, every participant independently locally trains AI models on their device. AI models are uploaded in decentralized storage and are recorded on the blockchain in an NFT format.
Participants are able to evaluate other uploaded models with their data and record the evaluation score on the chain. After the number of evaluations meets a certain threshold, the local model is a candidate for being aggregated into a global model.
Due to Blockchain's structure, records of compensation are released. In addition, participants will only participate in the study if they can be verified and deemed valid. Therefore, it is good to keep and manage the license of the temper-proof model and provide incentives by managing it in token form.
Through Federated Learning and Differential Privacy, participants are provided with appropriate compensation/ownership as much as they contributed to the model's learning. In addition, participation in learning is possible without infringement of privacy in Solai.
Through federated learning, local data is processed on client devices, so the central server does not collect data. This provides privacy benefits. Additionally, learning on client devices allows the model to learn from more diverse data compared to centralized learning, using distributed data. Federated learning can also effectively work on large datasets because learning is performed on local data.