Sheldon Abstract: Optimization and Learning in Novel Hardware

The successes of machine learning and rapidly expanding need for computing power have generated interest in specialized hardware for certain costly computational tasks.  Neuromorphic and analog circuitry are approaches that generally use uniform networks of elements (e.g. spiking neurons or memristors) to embed a computation in the dynamics of a circuit.  I will introduce two ongoing projects at LANL within this field: implementing backpropagation on the newest generation of spiking neural network hardware and embedding optimization problems in the dynamics of memristive circuits.

Refs: A Pulse-gated, Neural Implementation of the Backpropagation Algorithm
https://dl.acm.org/doi/10.1145/3320288.3320305

Asymptotic Behavior of Memristive Circuits https://doi.org/10.3390/e21080789