Page Comparison

Organizer:

Sumedh Risbud

Time:

9:00 am - 10:00 am (PT), Thu, Feb 11, 2021

Abstract:

Through this panel discussion, we wish to provide a novel perspective regarding neuromorphic hardware: as accelerators for solving large-scale optimization problems, with orders of magnitude gains in performance and energy. Following the idea of architecture-algorithm co-design, we will discuss how the architectural features of neuromorphic hardware enable efficient solutions of certain optimization problems as well as how the algorithmic features of certain optimization problems make them amenable to efficient neuromorphic solutions. We will build on state-of-the-art research of the panelists (e.g., mapping genetic algorithms, NP-hard problems like travelling salesman, and minimax dynamics onto spiking networks), in addition to the results obtained at Intel Labs on constraint satisfaction, graph search, and sparse coding.

Panelists:

• Wolfgang Maass (TU Graz) Wolfgang Maass

• Johan Kwisthout (Radboud University) Johan Kwisthout

• Udayan Ganguly (IIT Bombay) udayan@ee.iitb.ac.in

• Cengiz Pehlevan (Harvard University) Cengiz Pehlevan

• Ojas Parekh (Sandia National Lab)

• Gabriel Fonseca (Intel Labs, Neuromorphic Computing Lab)

• Prasad Joshi (Deactivated) (Intel Labs, Neuromorphic Computing Lab)

Pre-requisites/

co-requisites:

Attending Neuromorphic Solutions for Optimization Problems I will be useful, but not required

Recording:

Not available yetLink to recording

Link to Presentation:

Not available yetLink to slides

Previous INRC Forum Presentations by the Panelists:

• Gradient-based learning with Hebbian plasticity in deep and structured neural networks
  Cengiz Pehlevan

• Towards a neuromorphic complexity theory
  Johan Kwisthout