Xuan Zhang disputerer på kunstig intelligens - Universitetet i Agder
0
Hopp til hovedinnhold

Xuan Zhang disputerer på kunstig intelligens

Automatikk og datamaskiner har befridd mennesket fra tunge og repeterende arbeid. I økende grad blir datamaskinene smartere. Xuan Zhang forsker på automater som lærer gjennom respons.

Xuan Zhang disputerer med avhandlingen “Advances in the Theory and Applications of Estimator-based Learning Automata”. (Foto: Privat)

Xuan Zhang disputerer med avhandlingen “Advances in the Theory and Applications of Estimator-based Learning Automata”. (Foto: Privat)

I avhandlingen “Advances in the Theory and Applications of Estimator-based Learning Automata” forsker Xuan Zhang på en undergruppe av kunstig intelligens (AI) som blir forkortet LA – Learning Automata.

Xuan Zhang har fulgt doktorgradsprogrammet ved Fakultet for teknologi og realfag med spesialisering i IKT.

Slik beskriver kandidaten selv essensen i avhandlingen:

Advances in the Theory and Applications of Estimator-based Learning Automata

The substitution of automated and intelligent systems to do tasks that used to be conducted by humans, came with the invention of computers.

Made smarter to relieve people

This has developed, for the last decades, to become increasingly pervasive in our lives and in society. This, in turn, has triggered the need to make computers smart in order to relieve people, to the greatest extent, from laborious and repetitive tasks.

With the pervasiveness of computer systems, the study of Artificial Intelligence (AI) has developed into a field that tackles the challenges encountered by such systems to render them to be more intelligent and efficient.

As a subfield of AI, Learning Automata (LA) have been studied as a typical model for reinforcement learning, which, by interacting with a random environment, reinforces the ability of the computer system to make better decisions. In brief, an LA works in an iterative manner with its environment.

Learning by response

In each iteration, the LA chooses an action which triggers a response (reward or penalty) from the environment. Based on the response, and possibly on the history of responses, the LA adjusts its strategy of selecting an action in order to maximize the rewards that it would receive. A well-designed LA will be able to adapt itself to the optimal action offered by the environment through the process of learning.

Studies in the field of LA have been conducted for decades and the reported results and publications are numerous. Among the reported LA, the family of Estimator-based LA converges most rapidly to the optimal action by virtue of their ranking actions based on estimates.

In this thesis, we work to advance the theory and applications of Estimator-based LA, and our salient contributions are the following:

  • - New proofs for the convergence of Pursuit Algorithms.
  • - Design of Bayesian Pursuit Algorithms.
  • - Proof for the convergence of the BPAs.
  • - Application of the Bayesian Learning Automata in the field of Cognitive Radio Networks.

Disputasfakta:

Kandidaten:  Xuan Zhang obtained B.E. degree on Electronics and Information Engineering from Hunan University, China, in 2005. She obtained M.E. degree on Signal and Information Processing from Shandong University, China, in 2008. From May 2010 to January 2014, she worked as a Ph.D. research fellow in the field of Artificial Intelligence, in University of Agder, Norway. From September 2014 to December 2014, she worked in Teknova Company for the project of condition-based maintenance. Keywords of her research interests include: Machine Learning, Learning Automata, Stochastic Modeling and Optimization, Data Mining, Pattern Recognition and Classification, Condition-based Maintenance and Decision Support. 

Prøveforelesning og disputas finner sted i Rom C- 040, Campus Grimstad.

Professor Andreas Prinz, instituttleder ved Institutt for IKT, leder disputasen.

Tid for prøveforelesning: Torsdag 22. oktober 2015 kl 10:00

Oppgitt emne for prøveforelesning: “Recent Advances in Cognitive Radio Networks”

Tid for disputas: Torsdag 22. oktober 2015 kl 12:00

Tittel på avhandling: “Advances in the Theory and Applications of Estimator-based Learning Automata”

Søk etter avhandlingen i AURA - Agder University Research Archive, som er et digitalt arkiv for vitenskapelige artikler, avhandlinger og masteroppgaver fra ansatte og studenter ved Universitetet i Agder. AURA blir jevnlig oppdatert.

Opponenter:

Førsteopponent: Dr. Mohammad S. Obaidat

Annenopponent: Dr. Anders Kofod-Petersen

Bedømmelseskomitéen er ledet av professor Vladimir Oleshchuk, UiA

Veiledere i doktorgradsarbeidet var: Professor Ole-Christoffer Granmo (hovedveileder), professor John Oommen (bi-veileder).

Send studiet på mail