ABSTRACT:

When interacting with complex environments, humans can rapidly adapt their behavior in response to changes in task or context. To facilitate this adaptation, people often spend substantial periods of time contemplating possible futures before acting. In this talk, I will present empirical and modeling work exploring the critical balance between thinking and acting, and the factors affecting the content of our thoughts when we are making a decision. First, I’ll introduce a recurrent neural network model that learns when planning is beneficial, explaining variations in human thinking times and patterns observed in rodent hippocampal activity during navigation. Second, I’ll discuss how meta-learning enables neural networks to discover human-like planning strategies that blend elements of tree search and rollout algorithms. Finally, I’ll examine how working memory constraints shape reward encoding during sequential planning, revealing how humans strategically allocate cognitive resources based on decision relevance. Across these studies, recurrent neural networks emerge as powerful models for capturing the dynamic, iterative nature of human planning processes and their implementation in brain circuits

ABOUT THE SPEAKER:

Prof Marcelo Mattar is an Assistant Professor at New York University’s Department of Psychology and the Neuroscience Institute. He holds a PhD in Psychology from the University of Pennsylvania, where he also earned a Master’s in Statistics. His academic journey includes a Bachelor’s in Electronics Engineering from the Aeronautics Institute of Technology in Brazil. Prior to his current role, Prof. Mattar was a postdoctoral researcher at Princeton University under the mentorship of Nathaniel Daw and at the University of Cambridge’s Department of Engineering, working with Mate Lengyel.
His research integrates neuroscience, psychology, and computational modelling to explore how the brain supports flexible decision-making. Focusing on the interplay between learning, memory, and planning, Prof Mattar’s work aims to understand how the brain learns internal models from experience and simulates future scenarios to guide decision-making.