Dissertation > Excellent graduate degree dissertation topics show
Probabilistic reconstruction
Author: LiWeiHua
Tutor: LiuWeiYi
School: Yunnan University
Course: Communication and Information System
Keywords: Bayesian network Influence diagrams Probabilistic Local model Reduction Decomposition Merge
CLC: TP18
Type: PhD thesis
Year: 2010
Downloads: 70
Quote: 0
Read: Download Dissertation
Abstract
The explosion of knowledge and data flooding people need to dig out the useful information from the data, to obtain the required knowledge and reasoning knowledge. Random and uncertainty is an inherent feature of the real world itself. Therefore, we need a quantitative uncertainty assertion, and the assertion of these quantitative combine and supports uncertainty reasoning and decision analysis automated method. Probabilistic Bayesian networks and influence diagrams, becoming uncertain diagnosis, forecasting and decision support, classification, data mining mainstream paradigm, and to provide them with an intuitive, efficient, and reliable computing method. The probability Network is currently the most promising technologies in the application of artificial intelligence. Probabilistic knowledge representation and reasoning is also intelligent data analysis, knowledge discovery, as well as an important research direction uncertain artificial intelligence. The practical application problems often lead to a complex network of probability, probability network reasoning and learning exponential growth with the number of variables, and the needs of users are not static. Therefore, always ask for a model can effectively solve and it is more difficult to meet the changing needs of, or even impossible. In order to improve the model's applicability and efficiency of reasoning, the paper focuses on the problem of no data available Bayesian network and influence diagram structure optimization and model reconstruction. The main work and innovation, including the following three aspects. First, a the probability net reduction method. In this paper, two necessary and sufficient conditions to determine the characteristics of the local model graphics Bayesian network, given on the basis of a local model of learning methods. Secondly, an equivalence relation between defined herein settled point, and gathered by layerbylayer the equivalent node learning level model. Influence diagram, this paper defines a decision between the nodes of the decisionmaking collaboration and dependent relationship, and prove that it is an equivalence relation. Second, given the impact of the expected utility of the relevant decisionmaking set of variables that is relevant variables. In addition, based on the relevant decisionmaking and related variables to determine the impact of the local model of Figure method is given. In the present study, the local model of user concern submodels, and to ensure that local reasoning is equivalent to global reasoning. The hierarchical model provides a level of abstraction by encapsulating the details with partial complex problem. Reduction of the probability of network optimization model, the extended model availability, to avoid learning model from massive data, can also be reasoning is limited to a smaller model, the foundation to improve the efficiency of reasoning. Second, probability network biodegradable. Bayesian network give a simple expansion algorithm on the basis of previous work, The method is nondestructive to the Bayesian network can be decomposed into a set of models. Influence diagrams to solving a sequence collection expanded from a single decision node to the decision node defines a set of decision nodes extremum concept. Secondly, we prove that affect global optimal solution optimal solution determined by the local extreme value. Based on extreme concept, correlation analysis of the influence diagram given influence diagram decomposition method. Extremum set such that there is no standard for Solving the decomposition of the sequence of FIG possible and effectively expands the limitations of the correlation analysis. Probabilistic decomposition of the complex model into an equivalent set of submodels, to reduce the probability of network inference complexity. Third, the local fusion model. In this paper, two necessary and sufficient conditions to determine the global Probabilistic graphical features. On this basis, this paper presents a lossless fusion method does not rely on the data. Secondly, when the case of the inevitable conflict between the local model and inconsistent, the paper gives a method of learning the minimum I figure, it can be preserved as far as possible the individual information. In addition, we prove that the fusion method to learn directly from the data model. Fusion method can learn the data to the global model, to avoid subjectivity constructed by expert, makes reasoning more accurate, to complement probabilistic learning method,

Related Dissertations
 Experimental Research on Influence of Coal Reburning Reaction Conditions on NO_x Reduction by NonCatalytic Reduction of Ammonia,X701
 Application of Additives’ Drag Reducing Technology in District Heating System,TU995
 Experimentation Research for Deoxidize Reaction Process of NO_X Adopt Different Reburnning Fuels,TK227.1
 Study of Tin Alloy Carbon Composite Used for Lithium Ion Battery,TM912.9
 Query Processing and Optimization in Massive MultiDatabase Integration,TP311.13
 MultiSensor Information Fusion and Its Applications on Wearable Computer,TP202
 Research on MultiRobot Cooperative Pursuit Problem,TP242
 Research on Key Technologies of Automatic Steel Wire Galvanizing Production Line Based on Gas Reduction Method,TQ153.15
 Research and Implementation of the Conflict Resolution in Process Modeling of Magic Platform,TP311.5
 Physical Fields’ Characteristic Study & Structural Optimization of a SCR Denitrification Reactor,X773
 Study of Data Reduction Technique Based on Manifold Learning,TP311.13
 The Recognization of License Plate Based on EMD and Its Application,TP391.41
 Research on Performance of Merge and Acquisition of Flavor and Fragrance Industry,F271
 Minimally Invasive Locking Plate Fixation for Treatment of Pilon Fracture,R687.3
 Ir surface NH _{ 3 } catalytic decomposition and oxidation reaction mechanism studies,X131.1
 Based on Genetic Algorithm Pishihang irrigation canal water allocation marshalling model of,S274
 Highrise building under the jointed rock slope stability studies,TU457
 Calculation of Knowledge Granulation and Study of Its Application in Attribute Reduction,TP18
 TV shopping environment for consumers to reduce the perceived risk strategy study,G229.2F
 Study on Green Finance Promotes Industrial Restructuring,F832;F205
 The Model of Control and Model Calculations,O211.3
CLC: > Industrial Technology > Automation technology,computer technology > Automated basic theory > Artificial intelligence theory
© 2012 www.DissertationTopic.Net Mobile
