Translating cognitive models into neural and statistical descriptions of real-world multi-agent foraging behavior
Primary Area: applications to neuroscience & cognitive science
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: multi-agent systems, animal behavior, reinforcement learning, probabilistic methods, decision making
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Foraging is a multi-agent social behavior that has been studied from many perspectives, including cognitive science, neuroscience, and statistics. We start from a specific type of cognitive description -- agents with internal preferences expressed as value functions -- and implement it as a biologically plausible neural network. We also present an equivalent statistical model where statistical predictors correspond to components of the value function. We use the neural network to simulate foraging agents in various environmental conditions and use the statistical model to discover which features in the environment best predict the agent's behavior. Our intended primary application is the study of multi-species groups of birds foraging in real-world environments. To test the viability of the statistical approach, we simulate bird agents with different preferences, and use Bayesian inference to recover what each type of agent values. In the multi-agent context, we investigate how communication of information about reward location affects group foraging behavior. We also test our modeling technique on a previously published locust foraging dataset (Gunzel et al., 2023). After evaluating the effectiveness of our method on both synthetic and previously published data, we analyze new multi-agent foraging bird data we captured through high-resolution video recordings. Our method distinguishes between proximity preferences of ducks and sparrows within foraging groups. This analysis framework provides a principled, interpretable, and parametric approach for reasoning about how birds' preferences relate to their decisions about where to move in a complex multi-agent environment.
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: 6115
Loading