We recently developed a novel learning solution for unsupervised learning in robots based on resistive memory devices arranged in a crossbar fashion and validated it by navigating a robot in an unknown environment with randomly placed obstacles [1]. In this work, we study the effects of variations in device doping concentrations and the resistive states on the performance of the robot during navigation tasks. Such variabilities arise from the variation in process parameters during device fabrication. We have modeled the variabilities in the initial device doping concentration and in the update of the device resistive states. We have also considered the possibility of a device getting stuck in a low resistance state. This study will help us evaluate the performance of our learning scheme and develop specifications on acceptable range of variability in these devices for application-specific tasks.
Neuromorphic device specifications for unsupervised learning in robots
2017-06-01
737588 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Unsupervised shape learning in a neuromorphic hierarchy
British Library Online Contents | 2008
|Machine Learning With Neuromorphic Photonics
British Library Online Contents | 2019
Machine Learning With Neuromorphic Photonics
British Library Online Contents | 2019
Machine Learning With Neuromorphic Photonics
British Library Online Contents | 2019
Neuromorphic Learning From Noisy Data
NTRS | 1993
|