DIEP seminar by Fernando Rosas
Many systems exhibit nested emergent layers with their own rules and regularities, and our knowledge about them seems naturally organised around these levels. In this talk, I will put forward a formal approach to capture this correspondence, linking hierarchical structures in objective processes to Bayesian beliefs held by observers. For doing this, I will first present a framework to formalise the existence of self-contained levels via the concepts of informational, causal, and computational closure. Then, I will introduce a formal approach to world modelling, examining what it means to build an efficient internal representation of an perception-action loop. Finally, we will put these ideas together to identify equivariance of objective dynamics as a sufficient condition for the existence of multi-level beliefs. These ideas will be illustrated by a range of examples covering statistical mechanics and computational neuroscience.
Fernando Rosas is a lecturer in Computer Science and AI at the University of Sussex and Honorary Research Fellow Imperial College London. His work lies at the intersection of computational neuroscience, artificial intelligence, and complexity science.
He is interested in revealing the principles that drive the computations performed by distributed multi-agent systems:, which somehow allow them to become more than the sum of their parts — performing tasks collectively that couldn't be achieved individually, and even give rise to novel features that were not predefined beforehand.
If you wish to attend this seminar online, please send an email to r.lier@uva.nl to receive the zoom-link.