【作者】 王岩；

【导师】 周春光；

【作者基本信息】 吉林大学 ， 计算机应用技术， 2007， 博士

【摘要】 本文在全面分析和了解了生物信息学中操纵子预测和生物信号识别中味觉信号识别等的研究现状、研究热点和发展趋势的基础上,重点研究了操纵子预测和味觉信号识别的计算机识别算法。在将已有机器学习方法应用于操纵子预测和辅助搭建多种味觉信号识别系统模型基础上,提出了三种新型的可应用于操纵子预测和味觉信号识别的算法。本文的主要贡献和研究内容如下:(1)对生物信息学中的操纵子预测和生物信号识别中的机器味觉识别研究进行了全面综述。介绍了操纵子预测和机器味觉研究的产生背景、应用领域、研究现状、面临的挑战和发展趋势。将已有神经网络、支持向量机及遗传算法应用于操纵子预测中,辅助完成多种味觉信号识别系统模型。这些内容的讨论和分析是开展进一步研究的基础。(2)阐述了操纵子预测和机器味觉的相关机器学习基础理论。介绍了基本的多层前向神经网络的结构设计和学习理论、进化计算的基础理论以及概率分析和统计学习理论相关基础。(3)提出了一种最小不确定性神经网络味觉信号识别系统模型。(4)提出了一种边界法加速支持向量机的方法。(5)提出了一种量子群进化算法模型。本文的研究结果丰富了机器学习理论的应用研究,在概率分析与神经网络结合、神经网络的结构设计和参数学习、支持向量机学习和改进优化进化计算方法等方面的研究具有一定的理论意义和应用价值,为操纵子预测和机器味觉的实用化研究提供了有意义的方法和手段。更多还原

【Abstract】 Bioinformatics is an interdisciplinary approach drawing from specific disciplines such as biology, computer science and mathematics. It rised with human genome project at the end of 80’s. Because the current bioinformatics research is rooted in life science as well as microscopic molecular biology in structure research, the development of bioinformatics is mainly focusing on expanding the use of nucleotide and amino acid sequence data, including those to acquire, store, organize, archive, or analyze such data. Therefore during the process of explaining the essence of life, it is already known that not only the genetic information but also the expression regulation of genetic information is very important for organisms. So the construction of gene expression regulation network also becomes the key to discover the mystery of life. However as the regulation network research is too complexity and variety, scientists change their attentions to the individually sub-unit of regulation network which called operon. And furthermore they are trying to reconstruct the whole regulation network through the operon research.Similarly to bioinformatics, the macroscopic biological signals also are playing as important roles in our living, such as visual signal, sound signal, touch signal and so on. At present, vision sensation, hearing sensation and tactual sensation are all developed greatly in the robotics research domain. Some robots, which equipped with vision, hearing and touch abilities, have been used for practical purposes. Along with the development of life science and artificial intelligence research, the attempt of explore and imitate the sense function of taste and smell also has profound theory significance and wide spread application prospect. At present, Japan is in the top level of world in the research of taste and smell. But because involved so many disciplines, many crucial questions are still waiting for further studies.Though the research objects of operon prediction and taste signal recognition are different, the involved prediction and recognition algorithms of machine learning are similar. That provides the foundation for our research on machine learning algorithms in bioinformatics and biologic signal recognition.Machine Learning studies how to simulate human learning. It is converging from several sources, such as artificial intelligence, computational intelligence, statistics, mathematics, psychological, philosophy, adaptive control theory, informatics, biology etc. It nearly includes all human cognition domains. Fusing correlation machine learning methods, supplementing their superiorities, and then proposing new models and algorithms will promote the development of bioinformatics and biological signal recognition effectively.After comprehensively analyzing and understanding the present research status, opening topics and developing trendency in operon prediction of bioinformatics and taste signal recognition of biologic signal recognition, we mainly focus on the research of machine learning algorithms in operon prediction and taste signal recognition. And on the foundation of introduced known machine learning algorithms to operon prediction and assistanted to implement taste signal recognition system, we proposed three novel algorithms which may apply to operon prediction and taste signal recognition in the near future.(1) Introduced neuro network, support vector machine and evolutionary algorithm to operon prediction; assistanted to implement fuzzy neuro network based on voting strategy in taste signal recognition; used entropy clustering algorithm, class cover algorithm and rough sets algorithm to partition input space and extract rules respectively. All these works make the foundation of further research and study.(2) Introduced related basic machine learning theory in the operon prediction and the machine taste recognition, including multilayer feedforward neural network, evolution computation as well as statistics theory. The content has provided stabile theoretics and rationale for the application of our purpose.(3) A new model of taste signals recognition based on minimal uncertainty neural networks is proposed in this article. The model uses Bayesian Theorem and Particle Swarm Optimization (PSO) as tools to determine the parameters of the networks rapidly and efficiently. The identification of the taste signals of 10 kinds of tea is successful in utilization of this model. The experimental results show the feasibility and probability of this model to the identification of taste signals of tea.(4) We propose a boundary method to accelerate constructing the optimal hyperplane of support vector machines. The boundary, called key vector set, is an approximate small superset of support vectors set which extracted by Parzen window density estimation in the feature space. Experimental results on checkboard data and double helixs sets show that the proposed method is more efficient than conventional method and requires much less memory.(5) A new quantum swarm evolutionary algorithm (QSE) is presented, which is based on the quantum-inspired evolutionary algorithm (QEA). A new definition of Q-bit expression called quantum angle is proposed and an improved particle swarm optimization (PSO) is employed to update quantum angles automatically. The simulated results in solving 0-1 knapsack problem show that QSE is superior to QEA and many traditional heuristic algorithms. Meanwhile, the experimental results of 14 cities traveling salesman problem (TSP) show that it is feasible and effective for small-scale TSPs, which indicates a promising novel approach for solving TSPs.The research of this article has enriched the study of machine learning theory application. It has significance in applications, such as combination of probability analysis with neuro network, design and parameter study of neuro network structure, improvement and optimization of support vector machine, improvement of evolution computational etc. Furthermore, it provided significantthe method and strategy for the application of operon prediction and taste signal recognition.更多还原