A0401
Title: Additive-effect assisted learning
Authors: Jiawei Zhang - University of Kentucky (United States) [presenting]
Yuhong Yang - University of Minnesota (United States)
Jie Ding - University of Minnesota (United States)
Abstract: In an increasing number of machine learning applications, multiple learning agents hold datasets that can be collated by a particular identifier and have different features. These agents are often decentralized in nature and may appropriately assist each other in improving modeling performance. A two-stage assisted learning architecture is developed for an agent, Alice, to seek assistance from another agent, Bob, without sharing data. In the first stage, a privacy-aware hypothesis testing-based screening method is proposed for Alice to decide on the usefulness of the data from Bob in a way that only requires Bob to transmit sketchy data. Once Alice recognizes Bob's usefulness, Alice and Bob move to the second stage, where they jointly apply a synergistic model training procedure. Nontrivial theoretical analyses are provided to show that Alice can asymptotically achieve the oracle performance as if the training were from centralized data under appropriate settings. Simulation studies and real data demonstrations, including health condition prediction, image classification, and internet attack detection, show the encouraging performance of the proposed approach.
