When
Thematic Session 1: Design and Interaction (Monday, 14:15)
Abstract
Memristors are a broad class of two-terminal circuit elements with programmable resistance values modulated by the magnitude and polarity of an applied charge. The programmable resistance allows memristive devices to be used as neuromorphic components that combine memory and computation in a single volume, exhibiting switching behaviours from low resistance to high resistance states, volatile or non-volatile memory, and rich non-linear I-V (current-voltage) curve dynamics. Memristors have been utilised as artificial synapses in Spiking Neural Networks (SNNs), in machine learning architectures such as Reservoir Computing (RC) and could provide energy-efficient methods to improve data bandwidth and memory handling in embedded musical AI.
Building on work by Gale (2013), Kiefer (2014), and Braund (2019) I explore the use of memristive SNNs in musical AI. In particular, I detail use in machine learning for audio perception and generation with a focus on software implementation of memristor models using fractional derivatives and stochastic parameters such the hysteresis of the IV loop. I then discuss how this functionality can be implemented in hybrid analog-digital Spiking Neural Interfaces (SNNIs) via discussion of: (i) hardware/software co-design (ii) materials choice and methods to build IRL memristive circuits from cheap electrode and RS layer materials and (iii) future design possibilities for these interfaces in musical AI.
Bio
Davin Browner is a PhD candidate at the Robotics Laboratory at the Royal College of Art, London. His research is focussed on methods to fabricate memristive and resistive switching devices for robotics and bio-informational interfaces from readily available materials using low-cost fabrication methods. This exploration of technical design is pursued in combination with investigation of methods to integrate these devices into overall frameworks for machine learning such as Spiking Neural Networks (SNNs). He also holds a BA in Philosophy from Trinity College Dublin where he wrote a dissertation on neurophilosophy and linguistics. Davin has worked on collaborative projects including the design of open source acoustic monitoring devices for agriculture with the John Innes Centre and neuron modelling and device prototyping with the Ikeuchi Lab, University of Tokyo. As a musician he has performed in venues including Cafe Oto, Occii, and Salon des Amateurs and at festivals including Meakusma.