Fig. 3: Schematic diagram and simulation results of various training schemes.
From: Memristor-based analogue computing for brain-inspired sound localization with in situ training
a The distributions of update values of samples corresponding to various angles. b Schematic diagram of the multi-threshold-update scheme. As the updating levels increase, there will be more thresholds for ΔW classification to apply different numbers of pulses on the memristor device. c Flow chart of in situ training with the multi-threshold-update scheme. d Simulation results under various update levels and variation factors. The factor scales the updating variation of the devices. e The total energy consumption of memristor-based sound localization under various number of update thresholds.
