- Venue: Tutorial as part of the IJCAI'22 conference.
- Room name: Strauss 1.
- Date: July 25 2022 (Monday).
- Time: Afternoon session 1 and 2 (A1, A2).
- Registration: IJCAI'22 registration.
- Contact: For any questions, please email tutorial organizers.
Overview
This tutorial will provide an overview of recent research on adversarial learning in sequential decision-making settings. In particular, the tutorial will focus on adversarial attacks and defense mechanisms in the context of agents based on multi-armed bandits, reinforcement learning, and multi-agent interactions. The tutorial will tentatively cover the content listed below.
-
Introduction
- Primer to sequential decision-making: multi-armed bandits, reinforcement learning, multi-agent interactions, and game playing.
- High-level overview of how adversarial sequential decision-making differs from adversarial supervised learning.
- High-level overview of attack strategies and defense mechanisms.
-
Multi-armed bandits
- Optimal attack strategies under different models of feedback corruption and objectives.
- Recent works on designing robust algorithms, key challenges, and open problems.
-
Reinforcement learning
- Discussion of different learning paradigms (e.g., imitation learning, offline RL, and online RL) and how they crucially differ for adversarial attacks.
- Optimal attack strategies for test-time, training-time, and backdoor attacks.
- Optimal attack strategies under different models of data corruption and attack objectives.
- Recent works on designing robust algorithms, key challenges, and open problems.
-
Multi-agent interactions and game-theoretic considerations
- Attacks in multi-agent systems via controlling other agents and non-oblivious attacks.
- Utilizing game-theoretic tools for defense against non-oblivious attacks.
-
Practical considerations and discussion
- Case studies of security threats against learning agents.
- Developing benchmarking tools and datasets in adversarial sequential decision-making.
- Open discussion with the audience to promote cross-community collaborations.
Organizers
- Goran Radanovic. Max Planck Institute for Software Systems (Saarbrucken, Germany).
- Adish Singla. Max Planck Institute for Software Systems (Saarbrucken, Germany).
- Wen Sun. Cornell University (Ithaca, NY, USA).
- Xiaojin Zhu. University of Wisconsin-Madison (Madison, WI, USA).