Budget-bounded Incentives for Federated Learning

Date: 

Monday, October 18, 2021, 1:00pm to 2:30pm

Presenter:  Boi Faltings
Topic:  Budget-bounded Incentives for Federated

Fall 2021

The EconCS Group holds an Economics and Computer Science research seminar each semester.

Fall 2021 meetings are held at 1 - 2:30 PM on Fridays. Seminar Coordinators for Fall '21 are Srivatsa R Sai and Daniel Halpern.

Abstract: We consider federated learning settings with independent, self-interested participants. As all contributions are made privately, participants may be tempted to free-ride and provide redundant or low-quality data while still enjoying the benefits of the FL model. Known game-theoretic schemes for rewarding truthful data do not take into account novelty of data. This creates arbitrage opportunities where participants can gain rewards for redundant data, and the federation may be forced to pay out more incentives than justified by the value of the FL model. We show how a scheme based on {\em influence} can both guarantee that the incentive budget is bounded in proportion to the value of the resulting FL model, and that reporting data as accurately as possible is the dominant strategy of the participants. We next investigate what happens when the test data used for computing the influence is also elicited from participants. We show that if a portion of the testing data is of low quality, the incentive scheme will induce data collection with exactly the same proportion of low-quality data, and thus it cannot be used to improve data quality. This can be overcome by robust aggregators which however lose the feature of budget-boundedness.