Masters Theses
Date of Award
8-2022
Degree Type
Thesis
Degree Name
Master of Science
Major
Computer Science
Major Professor
James S. Plank
Committee Members
Garrett S. Rose, Catherine D. Schuman
Abstract
The TENNLab software framework enables researchers to explore spiking neuroprocessors, neuromorphic applications and how they are trained. The centerpiece of training in TENNLab has been a genetic algorithm called Evolutionary Optimization For Neuromorphic System (EONS). EONS optimizes a single population of spiking neural networks, and heretofore, many methods to train with multiple populations have been ad hoc, typically consisting of shell scripts that execute multiple independent EONS jobs, whose results are combined and analyzed in another ad hoc fashion. The Islands project seeks to manage and manipulate multiple EONS populations in a controlled way. With Islands, one may spawn off independent EONS populations, each of which is an “Island.” One may define characteristics of a “stagnated” island, where further optimization is unlikely to improve the fitness of the population on the island. The Island software then allows one to create new islands by combining stagnated islands, or to migrate populations from one island to others, all in an attempt to increase diversity among the populations to improve their fitness. This thesis describes the software structure of Islands, its interface, and the functionalities that it implements. We then perform a case study with three neuromorphic control applications that demonstrate the wide variety of features of Islands.
Recommended Citation
Zheng, Chaohui, "The Islands Project for Managing Populations in Genetic Training of Spiking Neural Networks. " Master's Thesis, University of Tennessee, 2022.
https://trace.tennessee.edu/utk_gradthes/6463