A research project on how biological and artificial collectives learn and adapt to find collective solutions to complex problems
Research Unit 1, SCIoI Project 11
The general objective of this project is to study dynamical adaptation and learning of biological and artificial collectives and how this contributes to collective solutions of complex problems outperforming individuals. We will study mechanisms of how individual agents within the collective acquire and share knowledge.
In particular, we will explore the role of partial, potentially conflicting or even erroneous, prior knowledge. The general scenario is a group of agents exploring a complex environment containing an a priori unknown number of targets (“food”). The targets themselves are not trivially accessible but are a) distributed in a complex environment, and b) need to be handled in a particular way to be successfully exploited. We are particularly interested in how collectives composed of individuals with partial prior knowledge (e.g. trained on different aspects of the foraging problem) combine their knowledge, and what the role of naive individuals without prior knowledge is.
On the analytical side, the main objective is to understand the benefits of grouping in terms of collective cognition. In particular, we want to explore two broad scenarios: 1) Situations where non-overlapping, partial knowledge of different individuals enhances collective performance, versus 2) cases where it impairs collective problem solving either because some prior knowledge is misleading or different priors leads to a stalemate (e.g. trained on different cues). Finally, we want to explore how partial knowledge (priors) interacts with possible persistent (fixed) heterogeneity within the collective (animal personalities) [BierbachNC].
On the synthetic side, methods from distributed and learning control theory will be extended and used to investigate the dynamics of collective learning. In particular, we will study (i) the interplay between prior knowledge, individual learning and knowledge transfer between agents, (ii) the relevance of the communication topology, and (iii) the effect of agent heterogeneity with respect to prior knowledge and learning strategies, dynamics and distortion of shared knowledge.
Formal methods developed on the synthetic side will be validated respectively improved using observations from two experimental testbeds – a biological one and a robotic one. Vice versa, methodological results will lead to a deeper understanding and to improved hypotheses on the analytical side.