Collaborative learning from an optimization and game theory view

Thumbnail

Event details

Date 01.09.2023
Hour 14:0016:00
Speaker Dongyang Fan
Location
Category Conferences - Seminars
EDIC candidacy exam
Exam president: Prof. Anne-Marie Kermarrec
Thesis advisor: Prof. Martin Jaggi
Co-examiner: Prof. Daniel Kuhn

Abstract
In this write-up, we consider a decentralized learning setting where multiple agents collaborate to enhance their respective local objectives in terms of test performance and convergence speed. We investigate methods for incentivizing agents to participate in collaboration and share truthful information, optimal collaborative strategies for agents, and the extent to which agents can derive benefits from such collaboration. These questions are addressed through a dual prism, encompassing both optimization and game-theoretical frameworks.

Background papers
- Mathieu Even, Laurent Massoulié, Kevin Scaman. On Sample Optimality in Personalized Collaborative and Federated Learning. Neurips 2022: https://proceedings.neurips.cc/paper_files/paper/2022/hash/01cea7793f3c68af2e4989fc66bf8fb0-Abstract-Conference.html
- Kate Donahue, Jon Kleinberg. Model-sharing Games: Analyzing Federated Learning Under Voluntary Participation. AAAI 2021.
https://ojs.aaai.org/index.php/AAAI/article/view/16669
- Florian E. Dorner and Nikola Konstantinov and Georgi Pashaliev and Martin Vechev. Incentivizing Honesty among Competitors in Collaborative Learning and Optimization. Arxiv 2023
https://arxiv.org/abs/2305.16272

 

Practical information

  • General public
  • Free

Tags

EDIC candidacy exam

Share