Scalable, Energy-Efficient Spiking Markov Decision Processes for Neuromorphic Computing
When
Where
Abstract: Neuromorphic computing is inspired by the parallel, energy-efficient computing capabilities of the brain and is well suited for AI/ML applications, but there are also non-cognitive applications that have a demonstrated advantage over Von Neumann architectures when implemented neuromorphically. A spiking random walk algorithm via discrete-time Markov chains has been shown to have neuromorphic scaling advantages, but beyond Markov chains/processes, little work has been done for Markov reward and decision processes. In this talk I will discuss the relevant literature and what I propose to do to implement scalable, energy-efficient spiking algorithms for Markov reward and decision processes on neuromorphic hardware.