Loading...

Developing Hierarchical Active Learning Method Framework for Complex Systems Analysis

Javadian, Mohammad | 2017

321 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 49839 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Bagheri Shouraki, Saeed
  7. Abstract:
  8. In recent decades, the science of studying complex systems has started to evolve and mature. Complex systems research is becoming ever more important in both the natural and social sciences. The study of mathematical complex system models is used for many scientific questions poorly suited to the traditional mechanistic conception provided by science. Examples of complex systems are Earth's global climate, organisms, the human brain, social organization, an ecosystem, a living cell, and ultimately the entire universe. Motivations for studying complex and self-organized systems can be somewhat divided between science, or attempts to understand such systems, and engineering, or attempts to design or manipulate such systems for our own benefit. Complex systems science suffers from the lack of a common formal framework for analysis. There are a number of reasons for this. Because Complex Systems Science is broader than physics, biology, sociology, ecology, or economics, its foundations cannot be reduced to a single discipline. Furthermore, systems which lie in the gap between the ‘very large’ and the ‘fairly small’ cannot be easily modeled with traditional mathematical techniques. Therefore, the scientists viewing complex systems in different ways. In other words, in order to analyze a complex system, due to its complex behavior, the way of analyzing the system entirely depends on our viewpoint of the system. In this dissertation, we intend to analyze complex systems by a data mining tool (clustering) and a soft computing technique (ALM). Complex systems have been characterized by a large number of variables (high-dimensional), nonlinearities, uncertainties and emergent behavior. When people are difficult to understand a complex system, simplification is often used to build a model. High-performance computers enable complex systems to be simulated with more variables. This leads to the computational paradigm of scientific research. Computational models provide a means for people to observe, analyze and understand the observed system. Models can be tested by data and improved to better reflect the nature. Therefore the nature of the big data is more or less the same as complex systems. A promising strategy to improve our understanding and simplify large complex systems is to reduce the complexity of the corresponding complex system, yet preserving its main characteristics. A natural approach to reducing the complexity is to simplify the systems by decreasing their size. Organizing information into clusters uncovers a relationship between objects. This relationship may give us insight into the internal structure of data; highlighting underlying rules and recurring patterns. It may also segment information effectively for visualization or as preprocessing step for reducing the dimensionality of information. Experiences in many fields also exhibit that the cluster analysis is a useful technique to find structures in a complex system. The idea underlying any clustering procedure is that the entries of a large data set can be organized and classified into a certain number of natural groups. Active Learning Method (ALM), is one of the powerful algorithms in soft computing and fuzzy logic systems. ALM is developed based on two hypotheses about human brain functionality. The first hypothesis claims that when human brain is confronted with a new learning problem, it tends to break down complex problems into some simpler sub-problems and the second hypothesis claims that information in the human brain is interpreted in pattern-like images by considering uncertainty rather than numerical quantities. It seems that due to existence of valuable concepts in ALM, this method can be considered as an appropriate base in order to develop a framework for analyzing the complex systems. In this dissertation, we first propose a low dimensional clustering algorithm based on Active Learning Method. Then we use the proposed algorithm as a weak learn in a consensus clustering and propose a hierarchical consensus clustering algorithm in order to construct a framework for analyzing and clustering data resulted from complex systems. Finally, we use this hierarchical high-dimensional clustering algorithm in the hierarchical active learning method framework in order to model the complex systems. We also propose a general clustering fuzzification algorithm which can be applied to all clustering algorithm in order to fuzzify their results, then we use this algorithm in our framework in order to determine the fuzzy membership function of the partitions
  9. Keywords:
  10. Complex System ; Clustering ; Active Learning ; Fuzzy Clustering ; Fuzzy Modeling ; Memristor ; Ensemble Learning

 Digital Object List

 Bookmark

No TOC