UMass Machine Learning and Friends Lunch | Main / Computational Aspects Of Brain Dynamics Experiments Models And New AI Approaches

Abstract:

Recent progress with high-resolution brain imaging techniques, including functional Magnetic Resonance Imaging (fMRI), Magnetoencephalography (MEG), Positron Emission Tomography (PET), Electro-Corticography (ECoG), and Electroencephalography (EEG), demonstrates an amazing vista on the complex spatio-temporal dynamics of cortical processes. Experiments show that the brain generates oscillatory neuronal activity at a broad range of frequencies and that these oscillations correlate with cognitive activity.

In this talk we summarize some key results of brain imaging experiments and introduce computational models of the observed brain oscillations and cognitive processing. We elaborate on a class of models, according to which brains are perceived as open thermodynamic systems converting sensory data into meaningful knowledge during repetitive phase transitions. Cortical phase transitions are viewed as neural correlates of higher cognition, which can be implemented in computers to develop new principles of intelligent computing and superior AI.

Bio: Robert Kozma is Visiting Professor at Computer Science, UMass Amherst, and Director of the Biologically-Inspired Neural & Dynamical Systems (BINDS) Lab. His research focuses on developing novel artificial intelligent system motivated by the operation of brains, see https://www.cics.umass.edu/person/kozma-robert.