Data Science Seminar by Ermin Wei: Flexible and Faithful Federated Learning and Unlearning Methods
Speaker:
Associate Professor of Electrical and Computer Engineering
Associate Professor of Industrial Engineering & Management Sciences
Northwestern University
Title:
Flexible and Faithful Federated Learning and Unlearning Methods
Abstract:
Federated learning enables machine learning algorithms to be
trained over decentralized edge devices without requiring the exchange
of local datasets. We consider two scenarios in this talk. In the
first scenario, we have cooperative agents running distributed
optimization methods. Current literature fails to capture the
heterogeneity in agents’ local computation capacities. We propose
FedHybrid as a hybrid primal-dual method that allows heterogeneous
agents to perform a mixture of first and second order updates. We
prove that FedHybrid converges linearly to the exact optimal point for
strongly convex functions. In the second scenario, we consider
strategic agents with different data distributions. We analyze how the
distribution of data affects agents' incentives to voluntarily
participate and obediently follow traditional federated learning
algorithms. We design a Faithful Federated Learning (FFL) mechanism
based on FedAvg method and VCG mechanism which achieves (probably
approximate) optimality, faithful implementation, voluntary
participation, and balanced budget. Lastly, we analyze an alternative
approach to align individual agent’s incentive to participate by
allowing unlearning option. We propose a multi-stage game theoretic
framework and study the equilibrium properties.
Speaker Bio:
Ermin Wei is an Associate Professor at the Electrical and
Computer Engineering Department and Industrial Engineering and
Management Sciences Department of Northwestern University. She
completed her PhD studies in Electrical Engineering and Computer
Science at MIT in 2014, advised by Professor Asu Ozdaglar, where she
also obtained her M.S. She received her undergraduate triple degree in
Computer Engineering, Finance and Mathematics with a minor in German,
from University of Maryland, College Park. Her team won the 2nd place
in the GO-competition Challenge 1, an electricity grid optimization
competition organized by the Department of Energy. Wei's research
interests include distributed optimization methods, convex
optimization and analysis, smart grid, communication systems and
energy networks and market economic analysis.