Skip to Main Content
 

Global Search Box

 
 
 
 

Files

ETD Abstract Container

Abstract Header

Fuzzy Cognitive Maps: Learning Algorithms and Biomedical Applications

Abstract Details

2015, PhD, University of Cincinnati, Engineering and Applied Science: Electrical Engineering.
Knowledge discovery from data is a challenging problem, especially for data that is noisy, high-dimensional or contains very few samples. Fuzzy cognitive maps (FCMs) could be used to address these challenges. FCM is a graph model with fuzzy node values and fuzzy edge weights. It emulates the imprecise but robust cognitive process of human experts in analyzing complex systems. FCMs have been applied to a wide range of applications. It is difficult to apply FCMs to areas with limited domain knowledge, because the initial FCMs are usually constructed by experts. Data-driven FCM learning algorithms could address this problem. However most of these algorithms were demonstrated to construct only small scale FCMs with less than 40 nodes and the accuracy of the constructed model is low. In this dissertation, we focus on developing FCM learning algorithms that can accurately construct FCMs for several hundreds of nodes and applying the proposed algorithms to the gene regulatory network inference problem and the brain magnetic resonance (MR) image analysis problem. The first part of this dissertation focuses on the developing new FCM learning algorithms. To improve the performance of the FCM learning algorithms, we first propose an algorithm based on ant colony optimization which outperforms the other FCM learning algorithms on small scale problems. Then we propose a novel decomposed problem formulation, in which the weights for each node are estimated separately. This decomposed approach is especially helpful to meta-heuristic algorithms because without explicitly decompose the problem into smaller ones, these algorithms will waste computational resource on exploring the different combinations of the decomposable weights. Several open questions are studied under the decomposed framework. First, we construct a new set of benchmark problems in order to address the lack of large scale benchmark problems. Second, we compare four FCM learning algorithms, including ant colony optimization (ACO), differential evolution (DE), particle swarm optimization (PSO) and tournament-based genetic algorithms (TGA). We suggest using TGA when data volume is relatively small and noisy level is high and using ACO or DE otherwise. Third, we studied the applicability of FCM learning algorithms to causal relation discovery problem by analyzing the correlation between the objective function values and the errors in the learned weight matrix. The second part of this thesis focuses on applying FCMs to medical domains. The first application is to infer gene regulatory networks (GRNs), which is an important problem in biomedical research because it could be used to understand the normal and abnormal functions of cells. We propose a sparse FCM learning objective function and demonstrate the FCM based approach can construct GRNs more accurately than the other approaches, including Bayesian networks and ordinary differential equations. The second application is to analyze brain magnetic resonance images. An accurate analysis method is important to the diagnosis and pathology study of neurological conditions. We first develop an algorithm based on support vector machine and scale invariant features, a 3D histogram of gradients feature, a novel two-level classification framework and a classification algorithm based on single layer FCMs trained by a gradient descent algorithm. Experiment results suggest the FCM approach can predict disease status more accurately.
Ali Minai, Ph.D. (Committee Chair)
Yizong Cheng, Ph.D. (Committee Member)
Karen Davis, Ph.D. (Committee Member)
Long Lu, Ph.D. (Committee Member)
Lawrence Mazlack, Ph.D. (Committee Member)
138 p.

Recommended Citations

Citations

  • Chen, Y. (2015). Fuzzy Cognitive Maps: Learning Algorithms and Biomedical Applications [Doctoral dissertation, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1423581705

    APA Style (7th edition)

  • Chen, Ye. Fuzzy Cognitive Maps: Learning Algorithms and Biomedical Applications. 2015. University of Cincinnati, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1423581705.

    MLA Style (8th edition)

  • Chen, Ye. "Fuzzy Cognitive Maps: Learning Algorithms and Biomedical Applications." Doctoral dissertation, University of Cincinnati, 2015. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1423581705

    Chicago Manual of Style (17th edition)