Jan Drugowitsch, PhD

Every decision and behavior is haunted by uncertainty, introduced by the noisy and ambiguous nature of the world that surrounds us. Despite this, we make such decisions with seeming ease. Our goal is to understand the fundamental computations and their neurobiological implementations that allow the nervous system to support such efficient behavior.

By using tools from machine learning and neuroscience, we approach this goal by developing theories about how networks of neurons are able to infer, represent, and process the state of the world, and how this processing leads to competent decisions. These theories are on one hand guided by statistical principles and the approximations required to keep the resulting computations tractable. On the other hand, they are constrained by and scrutinized in the light of our knowledge of the architecture of the nervous systems, and by observed behavior and neural activity.

Current research focuses on decisions based on perceptual evidence. We have previously shown that, in this context, decision-makers are able to trade off the time they contemplate such decisions with their accuracy in a close-to-optimal manner. This was demonstrated in rather restricted situations, and the current aim is to extend theories and collect behavioral evidence that pushes the boundary towards the realism of every-day decisions. We are further asking how decision strategies change once these decisions are between options of different intrinsic values, and what might be the computations involved to make up one’s mind about these values.

In the close future, we plan on extending our investigations not only to decisions that require statistical computations of higher complexity, but also to address behavior of higher dimensions, such as spatial navigation under uncertainty.