Neuromorphic computing utilizes electronic technology to replicate the computational properties of biological processes.
Spiking neural networks are compute engine of neuromorphic models. Unlike deep neural nets, neuronal outputs in spiking networks are time-space encoded binary values. Inspired by biological neural nets, neuromorphic chips can realize connections within arbitrary neurons.
In this talk, we provide an overview of AI technology and discuss how neuromorphic computing offers directions for AI chips in long term. Specifically, we present dynamic vision sensor (DVS), a neuromorphic device, and its applications to low-compute, low latency detection tasks.