Keywords: Meta-Learning, Videogames
TL;DR: The ways in which competitive videogame players adapt to slight changes in games can be thought of as a Meta-Learning problem.
Abstract: The field of meta-learning involves, not the training of a model for a particular task, but using training on a variety of related tasks to develop a model that transfers readily and generally to any of a set of similar tasks from some distribution. Similarly, top players of competitive videogames are required to contend with frequent updates or patches to their game of choice, and are expected to adapt to the changes quickly and without a significant decrease in skill. By considering each patch of a competitive game to be a separate task, drawn from a distribution describing versions thereof, the adaptation of players to new patches becomes analogous to the adaptation of models to new tasks. This paper seeks to describe the process of players adapting to updates of their games in the language of meta-learning, and then to analyze that process to inspire future modifications to the meta-learning paradigm.
Proposed Reviewers: Dr. Jane Wang, wangjane@google.com
Dr. Jennifer Raymond, jennifer.raymond@stanford.edu
0 Replies
Loading