dc.contributor.advisor |
Gahegan, M |
en |
dc.contributor.author |
Arvay, Adam |
en |
dc.date.accessioned |
2018-08-12T21:46:38Z |
en |
dc.date.issued |
2018 |
en |
dc.identifier.uri |
http://hdl.handle.net/2292/37637 |
en |
dc.description.abstract |
The work discussed in this thesis aims to improve the ability of humankind to understand complex systems by automating the model building process. These techniques fall within the discipline of computational scientific discovery, which has its roots in the early development of artificial intelligence. The approaches developed in this thesis differ from most existing automated model building techniques because they place an emphasis on discovering and expressing models in terms that human beings can easily understand. The primary problem addressed by this work is that many existing computational scientific discovery approaches are too computationally intensive. The proposed solution involves developing a new ratebased process modelling framework. It is based around the idea of using linear regression to estimate parameter values instead of gradient descent. The regression-based parameter estimation algorithm is implemented into a system called Regression-guided Process Modeller (RPM) that is tested on both empirical and synthetic data. Results from testing demonstrate that RPM is able to find models more than 83,000 times faster than the existing state-of-the-art system. This work goes on to describe two major additional improvements to the initial implementation as well as the challenge of creating synthetic data. The first improvement was implemented in the system called APM that allowed an existing known model to be used as the starting point to construct new models that explain unseen data. The second improvement in the system Selection-based Process Modeller (SPM) changed the search technique from a greedy technique to a backwards selection based heuristic. Empirical evidence is provided to demonstrate the effect of the improvements compared to the initial system RPM. Synthetic models and data are necessary to evaluate the discovery stems and their creation requires overcoming many of the challenges that computational scientific discovery is intended to automate. The new modelling framework, and the subsequent working discovery systems RPM, APM and SPM, provide a proof of concept that rate-based process modelling framework can find models more efficiently that previous approaches. These initial systems provide a starting point for the application of the ratebased process modelling framework as a computational scientific discovery tool to real-world dynamic systems. |
en |
dc.publisher |
ResearchSpace@Auckland |
en |
dc.relation.ispartof |
PhD Thesis - University of Auckland |
en |
dc.relation.isreferencedby |
UoA99265072813702091 |
en |
dc.rights |
Items in ResearchSpace are protected by copyright, with all rights reserved, unless otherwise indicated. Previously published items are made available in accordance with the copyright policy of the publisher. |
en |
dc.rights.uri |
https://researchspace.auckland.ac.nz/docs/uoa-docs/rights.htm |
en |
dc.rights.uri |
http://creativecommons.org/licenses/by-nc-sa/3.0/nz/ |
en |
dc.title |
Computational Scientific Discovery Using Rate-Based Process Models |
en |
dc.type |
Thesis |
en |
thesis.degree.discipline |
Computer Science |
en |
thesis.degree.grantor |
The University of Auckland |
en |
thesis.degree.level |
Doctoral |
en |
thesis.degree.name |
PhD |
en |
dc.rights.holder |
Copyright: The author |
en |
dc.rights.accessrights |
http://purl.org/eprint/accessRights/OpenAccess |
en |
pubs.elements-id |
751452 |
en |
pubs.record-created-at-source-date |
2018-08-13 |
en |
dc.identifier.wikidata |
Q112935525 |
|