We use cookies to ensure that we give you the best experience on our website. By continuing to browse this repository, you give consent for essential cookies to be used. You can read more about our Privacy and Cookie Policy.

Durham e-Theses
You are in:

Evolutionary computation based on nanocomposite training: application to data classification

VISSOL-GAUDIN, ELEONORE,GABRIELLE,BLANCH (2020) Evolutionary computation based on nanocomposite training: application to data classification. Doctoral thesis, Durham University.

PDF - Accepted Version


Research into novel materials and computation frameworks by-passing the limitations of the current paradigm, has been identified as crucial for the development of the next generation of computing technology. Within this context, evolution in materio (EiM) proposes an approach where evolutionary algorithms (EAs) are used to explore and exploit the properties of un-configured materials until they reach a state where they can perform a computational task. Following an EiM approach, this thesis demonstrates the ability of EAs to evolve dynamic nanocomposites into data classifiers. Material-based computation is treated as an optimisation problem with a hybrid search space consisting of configuration voltages creating an electric field applied to the material, and the infinite space of possible states the material can reach in response to this field. In a first set of investigations, two different algorithms, differential evolution (DE) and particle swarm optimisation (PSO), are used to evolve single-walled carbon nanotube (SWCNT) / liquid crystal (LC) composites capable of classifying artificial, two-dimensional, binary linear and non-linear separable and merged datasets at low SWCNT concentrations. The difference in search behaviour between the two algorithms is found to affect differently the composite’ state during training, which in turn affects the accuracy, consistency and generalisation of evolved solutions. SWCNT/LC processors are also able to scale to complex, real-life classification problems. Crucially, results suggest that problem complexity influences the properties of the processors. For more complex problems, networks of SWCNT structures tend to form within the composite, creating stable devices requiring no configuration voltages to classify data, and with computational capabilities that can be recovered more than several hours after training. A method of programming the dynamic composites is demonstrated, based on the reapplication of sequences of configuration voltages which have produced good quality SWCNT/LC classifiers. A second set of investigations aims at exploiting the properties presented by the dynamic nanocomposites, whilst also providing a means for evolved device encapsulation, making their use easier in out-of-the lab applications. Novel composites based on SWCNTs dispersed in one-part UV-cure epoxies are introduced. Results obtained with these composites support their choice for use in subsequent EiM research. A final discussion is concerned with evolving an electro-biological processor and a memristive processor.
Overall, the work reported in the thesis suggests that dynamic nanocomposites present a number of unexpected, potentially attractive properties not found in other materials investigated in the context of EiM.

Item Type:Thesis (Doctoral)
Award:Doctor of Philosophy
Keywords:evolution in materio, evolutionary algorithms, data classification, optimisation, differential evolution, particle swarm optimisation,nanocomposites, single-walled-carbon-nanotubes, liquid crystals, UV-cure epoxy
Faculty and Department:Faculty of Science > Engineering, Department of
Thesis Date:2020
Copyright:Copyright of this thesis is held by the author
Deposited On:15 May 2020 11:55

Social bookmarking: del.icio.usConnoteaBibSonomyCiteULikeFacebookTwitter