Computing with self-organised networks
A research forum to discuss performing computations with self organised networks.
Join us for a forum where we are discussing computating with self-organised networks. We are join by three fantastic speakers outline below and have Q&As for each speaker so that we can hear from you.
Neuromorphic dynamics in nanowire networks
Speaker: Prof. Zdenka Kuncic (University of Sydney)
Polymer-coated metal nanowires self-assemble into a complex network with heterogenous interconnectivity and memristive cross-point junctions. Under electrical stimulation, the interplay between memristive junctions and complex circuitry results in emergence of collective neural-like dynamics (e.g. long-range transport, switch synchronisation, criticality). As the hardware already embeds neural network-like circuitry, artificial neural network algorithms do not need to be implemented to achieve learning. Instead, a reservoir computing approach can be exploited to demonstrate efficient learning that harnesses the adaptive dynamics of the neural-like network for temporal information processing.
Synaptic plasticity effects in self-organized memristive nanowire networks
Speaker: Dr. Gianluca Milano (INRiM – Politecnico di Torino)
Self-organized memristive nanowire (NW) networks realized with a bottom-up approach represent promising platforms for the implementation of unconventional types of computing paradigms. Differently from conventional memristive crossbar architectures realized with a top-down approach where the behavior depends on each single memory element that has to be addressed independently, NW networks exhibit an emergent behavior where players are not individual nano-objects but their interactions. The NW network emergent behavior is related to i) wiring plasticity due to the rupture/rewiring of single NWs and ii) weight plasticity associated to resistive switching phenomena at NW junctions. Thanks to these effects, the NW network reconfigures when subjected to external electrical stimuli showing structural plasticity with heterosynaptic capability.
Dopant Network Processing Units
Speaker: Dr. Hans-Christian Ruiz Euler (University of Twente)
The exponential growth in the computational demands of deep learning require novel hardware designs. In this talk, I will introduce a promising novel approach to neural information processing using “Dopant Network Processing Units” (DNPUs). Each of these tuneable, highly non-linear nano-electronic devices can solve many linearly non-separable binary classification problems. By using DNPUs as novel neurons with high computational capacity, I will show how this technology can pave the way towards material-based neural networks that offer efficient information processing with low latency and energy consumption.