Mechanisms that Incentivize Data Precision in Centralized Learning

In centralized machine learning, the quality of the data contributed by agents significantly impacts model performance. Without proper mechanisms in place, some agents may choose to contribute less and rely on the efforts of others. This work introduces an algorithm to detect variance among agents and further extends a mechanism to incentivize them to enhance the quality of their data contributions based on the variance detection ability, where agents will be rewarded according to the quality of their contributions. Our approach builds upon previous research that primarily focused on maximizing the quantity of collected samples, extending it to explicitly address data quality. The mechanism we extend is shown to maximize the quality of data contributed by the agents at equilibrium.

M.Sc. student under the supervision of Prof. Kfir Yehuda Levy.