Collective Information Processing Romanczuk Lab

From a chemical engineering undergraduate degree, to a biophysics/biochemistry MSc at KU Leuven with a dynamical systems thesis and a paper on simulating task partitioning in ants, Patrick has been pursuing a PhD in the CIP lab via a self-secured grant from the Elsa-Neumann-Stipendium.

By training artificial neural network (ANN) controlled agents to solve a classic navigation task, Patrick explores fundamental tradeoffs between different informational input qualities used to construct effective visual navigational strategies: between angle and distance information at the individual level, and between spatial and social information at the collective level. At the individual level, visual agents without memory solve the Morris Water Maze, demonstrating the validity of a sequential route-based algorithm, distinct from both response-based and map-based explanations. At both individual and collective levels - a diversity of strategies emerge, where characterizing behavioral variation sheds light on fundamental affordances and tradeoffs inherent in visual navigation. Finding a gap between existing reinforcement learning (RL) approaches and the cognitive behavioral study of individual differences, he and colleagues put together a perspective piece on how to augment RL as a tool to study the bottom-up emergence of individuality.

Search for Patrick Govoni's papers on the Publications page