Generative Modeling of Individual Behavior at Scale
Primary Area: visualization or interpretation of learned representations
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: parameter efficient finetuning, chess, play style, stylometry, interpretation of learned representations
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Recent years have seen a growing interest in using AI to model human behavior, particularly in domains where humans learn from or collaborate with this AI. While most existing work attempts to model human behavior at an aggregate level, our goal is to model behavior at the individual level. Recent work in the domain of chess has shown that behavioral stylometry, or the task of identifying a person from their actions alone, can be achieved with high accuracy among a pool of a few thousand players. We provide a new perspective on behavioral stylomery by connecting it to the vast literature of transfer learning in NLP. Specifically, by casting the stylometry problem as a multi-task learning problem---where each task is a distinct person---we show that parameter efficient fine-tuning (PEFT) methods can be adapted to perform stylometry at an unprecedented scale (47,864 players), while enabling few-shot learning for unseen players. Our approach leverages recent modular PEFT methods to learn a set of skill parameters that can be combined in different ways using style vectors. Style vectors enable two important capabilities. First, they make our approach generative, in that we can generate actions in the style of a player by simply indexing into that player's style vector. Second, they induce a latent style space that we can interpreted and manipulated algorithmically. This allows us to compare different player styles, as well as synthesize new (human-like) styles, e.g., merging the styles of two players or interpolating between their styles.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7455
Loading