On the use of collaborative interactions for embedded sensing applications: Memristor networks as intelligent sensing substrates

Paper i proceeding, 2021

A novel sensing approach has been investigated in which environment-sensitive memristor networks are used as intelligent sensing substrates. A substrate collects pieces of environment-related information over time and encodes this information into its state. The stored information can be extracted by monitoring how the substrate responds to an external drive signal. An advantage of this indirect sensing approach is that the drive signal can be optimised to make the inference process efficient: even small pieces of information (which might go unnoticed in the traditional sensing setup) are collected. To demonstrate the main ideas an instance of a binary classification problem has been investigated. A separability index has been used as a measure of the substrate quality. By simulating the dynamics of a large number of memristor networks and computing their separability indices, it has been found that heterogeneous networks with delayed feedback elements make good sensing substrates.