(A) hopfield-like artificial neural network for solving inverse radiation transport problems = 역 방사선 수송 문제 해를 위한 Hopfield형 인공신경망

In this thesis, we solve inverse radiation transport problems by an Artificial Neural Network(ANN) approach. ANNs have many interesting properties such as nonlinear, parallel, and distributed processing. Some of the promising applications of ANNs are optimization, image and signal processing, system control, etc. In some optimization problems, Hopfield Neural Network(HNN) which has one-layered and fully interconnected neurons with feed-back topology showed that it worked well with acceptable fault tolerance and efficiency. The identification of radioactive source in a medium with a limited number of external detectors is treated as an inverse radiation transport problem in this work. This kind of inverse problem is usually ill-posed and severely under-determined; however, its applications are very useful in many fields including medical diagnosis and nondestructive assay of nuclear materials. Therefore, it is desired to develop efficient and robust solution algorithms. Firstly, we study a representative ANN model which has learning ability and fault tolerance, i.e., feed-forward neural network. It has an error backpropagation learning algorithm processed by reducing error in learning patterns that are usually results of test or calculation. Although it has enough fault tolerance and efficiency, a major obstacle is "curse of dimensionality"--required number of learning patterns and learning time increase exponentially proportional to the problem size. Therefore, in this thesis, this type of ANN is used as benchmarking the reliability of the solution. Secondly, another approach for solving inverse problems, a modified version of HNN is proposed. When diagonal elements of the interconnection matrix are not zero, HNN may become unstable. However, most problems including this identification problem contain non-zero diagonal elements when programmed on neural networks. According to Soulie et al., discrete random iterations could produce the stable minimum state of an...
Advisors
Cho, Nam-Zinresearcher조남진researcher
Publisher
한국과학기술원
Issue Date
1997
Identifier
114544/325007 / 000953428
Language
eng
Description

학위논문(석사) - 한국과학기술원 : 원자력공학과, 1997.2, [ iv, 38 p. ]

Keywords

Hopfield neural network; Artificial neural networks; Inverse radiation transport problems; Optimization; 최적화; Hopfield 신경망; 인공신경망; 역 방사선 수송 문제

URI
http://hdl.handle.net/10203/49371
Link
http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=114544&flag=t
Appears in Collection
NE-Theses_Master(석사논문)
Files in This Item
There are no files associated with this item.
  • Hit : 83
  • Download : 0
  • Cited 0 times in thomson ci

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0