Meta-Referential Games to Learn Compositional Learning Behaviours

Kevin Yandoka Denamganai, Sondess Missaoui, James Alfred Walker

Keywords: referential game, language grounding, compositionality, systematicity, few-shot learning, meta-learning, reinforcement learning, language emergence, symbolic behaviours, benchmark

TL;DR: We present a novel framework of Meta-Referential Games upon which we built a novel benchmark, the Symbolic Behaviour Benchmark, and use it in an initial study to evaluate AIs' abilities to learn compositional learning behaviours.

Abstract:

Human beings use compositionality to generalise from past to novel experiences, assuming that past experiences can be decomposed into fundamental atomic components that can be recombined in novel ways.
We frame this as the ability to learn to generalise compositionally, and refer to behaviours making use of this ability as compositional learning behaviours (CLBs).
Learning CLBs requires the resolution of a binding problem (BP).
While it is another feat of intelligence that human beings perform with ease, it is not the case for artificial agents.
Thus, in order to build artificial agents able to collaborate with human beings, we develop a novel benchmark to investigate agents’ abilities to exhibit CLBs by solving a domain-agnostic version of the BP.
Taking inspiration from the Emergent Communication, we propose a meta-learning extension of referential games, entitled Meta-Referential Games, to support our benchmark, the Symbolic Behaviour Benchmark (S2B).
Baseline results and error analysis show that the S2B is a compelling challenge that we hope will spur the research community to develop more capable artificial agents.

Primary Area: datasets and benchmarks

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.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

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: 11749