CS Events

Computer Science Department Colloquium

Intelligence beyond the neuron: Energy-Efficient and Robust Brain-inspired Computing Algorithms Evaluated in Neuro-Morphic and Neuro-Rehabilitation Robots

 

Download as iCal file

Monday, September 14, 2020, 10:30am

 

Speaker: Dr. Konstantinos P. Michmizos

Bio

Konstantinos P. Michmizos is an Assistant Professor of Computer Science, an Executive Council Faculty at Center for Cognitive Science, and a Faculty member of the Brain Health Institute and CBIM at Rutgers University. Previously he held positions at Harvard Medical School and MIT.

Location : Via Webex meeting

Event Type: Computer Science Department Colloquium

Abstract: In our strive for ever more efficient and intelligent systems, the Von Neumann computing paradigm that has been serving us for fifty years is no longer sufficient. The problem lies in the paradox that, despite the pioneering early connections between computation and brain studies, today’s algorithms and systems diverge from what we have lately learned about the brain’s information processing and learning. The neuronal and non-neuronal computing principles lead to dramatic efficiencies in nature but make them extremely ill-suited for executing on today’s energy- and memory-greedy processors. This talk will present our efforts to advance non-Von Neumann computations that draw from the brain’s functional analogies and redefine algorithms as spiking neuronal astrocytic networks, where memory, learning and computing are tightly integrated. Specifically, it will describe how we a) develop and synthesize computational models and learning algorithms grounded in recent breakthroughs in neuroscience, b) translate this spectrum of knowledge into scalable computing primitives and deploy them on large-scale neuromorphic processors, and c) evaluate the emerging behavior as a solution to foundational problems in robotics, leading to 2 orders of magnitude more energy-efficient and significantly more robust solutions compared to the state-of-the-art deep learning. The talk will conclude with the learned lessons and the untapped challenges in harnessing intelligence for robots, by either archetyping it to serve people’s life or targeting it to assist people’s recovery.

Contact  Host: Dr. Matthew Stone

The recording of the talk can be found at this link

https://rutgers.webex.com/recordingservice/sites/rutgers/recording/playback/b6ecf8a974894d5a8e2f648adb888249

Password: DtPjYPv3