BEGIN:VCALENDAR METHOD:PUBLISH PRODID:-//Apple Computer\, Inc//iCal 1.0//EN X-WR-CALNAME;VALUE=TEXT:USC VERSION:2.0 BEGIN:VEVENT DESCRIPTION:Speaker: Ben Dongsong Huh, Postdoctoral Fellow, Computational Neurobiology Laboratory, Salk Institute for Biological Studies Talk Title: Investigating the spike-based computations of the brain Abstract: Our brain uses temporal dynamics of neural activities to perform real-time computations: it processes time-varying streams of information and produces action sequences. How the brain coordinates the complex biophysical dynamics to form the basis for computation is a central problem in neuroscience. I apply optimal control theory to investigate how functionality of dynamical systems arises from first principles and, more specifically, to establish a unifying framework for understanding the dynamics and computations of the brain. The most prominent characteristics of biological neural networks is spikes: The brief impulse signals link individual neural dynamics and provide a unified currency for the asynchronous information processing in the brain. However, neuroscience lacks the theoretical framework for modeling how spikes represent information and perform computations in distributed network architectures. To solve this problem, I derived the first general learning algorithm for spiking neural networks from an optimal control principle, representing the first step in harnessing the computational potential of spikes. The spike-based computational principles can be extracted by analyzing how a trained network solves the computational tasks. More generally, this method allows combining the top-down deep learning approaches with the biophysical network properties to yield detailed models of neural systems that are both structurally and functionally accurate.\n This research has a wide range of engineering applications, including spike-based deep learning for neuromorphic devices, and next generation Brain-Machine-Interface and neuro-prosthetics that directly use spike signals for fine control. Most importantly, I aim to promote close collaborations between neuroscience and artificial intelligence research by providing a common theoretical framework. Host: Ellis Meng, PhD SEQUENCE:5 DTSTART:20180205T130000 LOCATION:DRB 145/145A DTSTAMP:20180205T130000 SUMMARY:Biomedical Engineering Department Guest Speaker UID:EC9439B1-FF65-11D6-9973-003065F99D04 DTEND:20180205T140000 END:VEVENT END:VCALENDAR