When
Where
Student: Sarah Luca
Title: Challenges and strategies in neuromorphic algorithm development: a comparative study across the stacks
Advisors: Misha Chertkov (Department of Mathematics) and Felix Wang (Sandia National Laboratories)
Location: Math Building, Room 402 | https://arizona.zoom.us/j/87830188626
Abstract: With the explosive rise in AI, scientific and edge computing applications in recent years, along with the approaching end of Moore’s law, there is a growing need for more efficient approaches to computing. Neuromorphic computing is an emerging computing paradigm inspired by the brain which can perform highly distributed, parallel computations in dynamically changing noisy environments in an extremely energy efficient way, making it an appealing alternative approach to computing for these applications. Due to its emerging nature and substantial differences in hardware design to conventional computing architectures, neuromorphic algorithms are challenging to develop. In this talk, I will discuss the current challenges and strategies to neuromorphic algorithm development at several levels across the stack: physical hardware implementation (digital versus analog), implementation approach (hardware constrained versus hardware aware), and parameter determination (manual versus machine learning); in the context of two different applications: solving Markov decision processes with value iteration, and performance of inference tasks in extreme conditions at the edge. In addition to highlighting challenges and strategies, this work contributes to the growing library of algorithms for neuromorphic computing and demonstrates how a brain-inspired approach can provide solutions to problems in real world applications.