Training Procedure
Anybody can propose any task one wants to train, and the neural network model structure and hyperparameters to use when training.
The proposal passed by the overall community begins the training procedure.
Repeat train round (Single Train Round Below)
Every participant independently and locally trains AI models on their device.
Participants upload AI models in decentralized storage, mint an NFT that leads to the model URL, and then commit to the training procedure.
Evaluation of the model is done by peer evaluation by other participants.
Local models become a candidate for the global model aggregation if the number of evaluations meets a certain threshold
Candidate Local models are aggregated and then proposed for the committee to be judged whether to pass the global model proposal.
If passed, commit tokens are given to the committing participants.
Last updated
