Task Allocation

Inference tasks can be allocated to service providers in exchange for tokens. Specifically, this model involves users exchanging tokenized tasks or requests in the form of digital assets, which can serve as a tradable unit of value for the execution of a specific machine learning model by service providers. Machine learning service providers can participate in the network by offering their computational resources to run inference on an input, and they are rewarded with tokens in exchange.

A significant challenge in designing a task-allocation system is that determining the optimal task allocation solution is NP-Hard [4]. We implement the differential privacy-based combinatorial double auction algorithm proposed by Zhai et al. between users and service providers, and trades are executed when the bid price of a user matches the seller’s ask price