Sensing-to-Learn and Learning-to-Sense: Principles for Designing Neuromorphic Sensors

Neurobiological systems have evolved over a billion years and serve as a good template for some text engineers to mimic when designing intelligent sensors and systems. For instance, neurobiological systems exploit noise and system non-linearity as a computational aid to push the limits of performance and energy efficiency. In contrast, in man-made technologies, these artifacts are generally considered to be a nuisance. This chapter's focus is on the neuromorphic concept of "sensing-to-learn" and "learning-to-sense," which are grounded in key neuromorphic adaptation principles based on noise exploitation and non-linear sensory processing techniques. "Noise shaping" and "jump-resonance" are two techniques that can extract salient sensing cues by exploring the synergy between noise and system non-linearity. We illustrate these concepts in the context of auditory and olfaction pathways, and we argue how these principles can be used to design the next generation of neuromorphic sensory interfaces.