Description:

All biological systems are learning systems. Moreover, all biological systems are physical systems. Therefore, there exists a direct connection between physics and learning. To tease out these statements, we will discuss efforts to uncover new learning mechanisms, or algorithms, from a physics perspective, aka physical learning, such as multi-mechanism learning (MML) and frequency propagation (FP).  MML can be implemented to recognize static patterns as well as dynamic ones and provides a simple physical explanation for the underlying basis of bidirectionality observed in neuronal circuits, for example. While MML and FP are supervised physical learning algorithms, efforts are also underway to develop unsupervised physical learning algorithms given the recent observation that physical learning can be viewed as feedback-based aging in a glassy landscape. All in all, these combined efforts point towards a unified framework of learning with and without brains that may help us build both better artificial intelligence systems and biological intelligence systems, including a network of brain organoids.  

Jennifer Schwarz, Professor, Syracuse University

Host: Istvan Kovacs