Call for Papers - Gamification and Multiagent Solutions Workshop - ICLR 2022

299 views
Skip to first unread message

Ian Gemp

unread,
Jan 24, 2022, 7:15:01 PM1/24/22
to Machine Learning News
We invite submissions to the first instance of the Gamification and Multiagent Solutions workshop hosted at ICLR 2022.

TL;DR: Modern machine learning is dominated by an optimization-first, single agent
approach. Can we revisit machine learning with a multiagent approach in mind? What problems can we pose as games among interacting agents?

Workshop description: Many of life’s intelligent systems are multiagent in nature: from market economies to ant colonies, from forest ecosystems to decentralized energy grids, intelligence is often a property of the whole, not its parts. These real-world examples suggest some deeper mathematical principle of intelligence, one grounded in games and multiagent interactions. However, modern machine learning primarily takes an optimization-first, single agent approach.

What do multiagent systems have to offer in the way of solutions? Generative adversarial networks (Goodfellow et al, 2014) reformulate learning a generative model as a two-player, zero-sum game. Similarly, EigenGame (Gemp et al, 2021) reformulates top-k singular value decomposition / principal component analysis as a k-player, general-sum game.  What other learning problems can we “Gamify” by casting them as games among interacting agents? What might we learn from reformulating machine learning from the ground up with a multiagent approach in mind?

Multiagent designs are typically distributed and decentralized which leads to robust and parallelizable learning algorithms. Interactions between multiple agents also drive the creation of curricula that challenge learning agents to improve generalization performance.

We want to bring together a community of experts to explore:
  • What makes a problem amenable to a multiagent approach?
  • Which natural multiagent systems exhibit intelligent behaviors we can reuse in artificial agents?
  • How do we shepherd systems of adaptive agents to useful equilibria?
  • Can we develop novel multiagent solutions to machine learning problems?
  • In which cases are multiagent approaches crucial to advancing state-of-the-art?
  • What new solutions will we find at the fixed points, equilibria, or attractors of our games that were not at the bottom of our loss functions?
By exploring this direction, we might gain a fresh perspective on machine learning, and unearth a new and exciting direction to build multiagent solutions.

Website with full details:

Timeline for submissions:
Submission Deadline: February 26, 2022 (AoE).
Acceptance Notification: March 26, 2022.
Camera Ready: April 16, 2022 (AoE).
Workshop: April 29, 2022.

Confirmed Speakers:
Constantinos Daskalakis, MIT
Sarit Kraus, Bar-Ilan University
David Parkes, Harvard
Lillian Ratliff, University of Washington
Elad Schneidman, Weizmann Institute of Science


Emailgamifica...@gmail.com

Ian Gemp (DeepMind) on behalf of the workshop organizers
Reply all
Reply to author
Forward
0 new messages