Participants

Publisher

The Publisher is the company, protocol, DApp, or individual that wants to utilize an AI model, either for internal use or as part of a user-facing product.

Trainer

Trainers are users and end-devices willing to participate in the collaborative training process. They are rewarded based on their contribution to the overall accuracy of the model. Typically, trainers are actual users of the Task Publisher, making them well-suited to train the Publisher’s model. However, there are scenarios where Task Publishers issue a public task open for training by non-users.

Aggregator

Aggregators are entities responsible for evaluating the Trainers' local training results and aggregating the model improvements proposed by the Trainers. Depending on the task, Aggregators may be a subset of the Trainers themselves who possess sufficient computational resources and have registered to serve both as Aggregators and Trainers, thus earning higher rewards. Alternatively, Publishers may permit any entity with adequate resources to register as an Aggregator for their tasks. This allows individuals with sufficient computing power to participate in the network and support its decentralization. Aggregators must stake the network’s native token and receive rewards or penalties based on their behavior and contributions.

Delegators

Aggregators are delegated tokens by Delegators, increasing their stake in the network and earning a percentage of its rewards.