Automatic speech recognition (ASR) by machine has received a great deal of attention in past decades. Speech recognition algorithms based on the Mel frequency cepstrum coefficient (MFCC) and the hidden Markov model (HMM) have a better recognition performance compared with other speech recognition algorithms and are widely used in many applications. In this thesis a speech recognition system with an efficient MFCC extraction algorithm and multi-mixture models is presented. It is composed of two parts: a MFCC feature extractor and a HMM-based speech decoder. / For the HMM-based decoder of the speech recognition system, it is advantageous to use models with multi mixtures, but with more mixtures the calculation becomes more complicated. Using a table look-up method proposed in this thesis the new design can handle up to 16 states and 8 mixtures. This new design can be easily extended to handle models which have more states and mixtures. We have implemented the new algorithm with an Altera FPGA chip using fix-point calculation and tested the FPGA chip with the speech data from the AURORA 2 database, which is a well known database designed to evaluate the performance of speech recognition algorithms in noisy conditions [27]. The recognition accuracy of the new system is 91.01%. A conventional software recognition system running on PC using 32-bit floating point calculation has a recognition accuracy of 94.65%. / In the conventional MFCC feature extraction algorithm, speech is separated into some short overlapped frames. The existing extraction algorithm requires a lot of computations and is not suitable for hardware implementation. We have developed a hardware efficient MFCC feature extraction algorithm in our work. The new algorithm reduces the computational power by 54% compared to the conventional algorithm with only 1.7% reduction in recognition accuracy. / Han Wei. / "September 2006." / Adviser: Cheong Fat Chan. / Source: Dissertation Abstracts International, Volume: 68-03, Section: B, page: 1823. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2006. / Includes bibliographical references (p. 108-111). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Electronic reproduction. [Ann Arbor, MI] : ProQuest Information and Learning, [200-] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstracts in English and Chinese. / School code: 1307.

在生物信息學中， 基因調控網絡推斷不斷受到人們的重視。各種不同的網絡模型被用來描述基因之間的調控關係， 其中包括布爾網絡， 概率布爾網絡， 貝葉斯網絡等。本文主要是討論基於數據的布爾網絡推斷。現在已經有很多方法來推斷節點是離散變量的網絡結構。比如REVEAL算法，Best Fit Extension 算法是兩種比較受歡迎的推斷網絡結構方法。並且他們在網絡的節點數目不是很多的情況下有很好的表現。然而， 現今很多方法對噪音和模型的不確定性沒有足夠的考慮。這也使得這些方法在實際應用中的表現不是很令人滿意。本文中， 我們用完全貝葉斯的方法去研究概率布爾網絡空間。在給定樣本的情況下， 我們提出了一種新的基於馬爾科夫鏈蒙特卡羅的算法。這種算法使得不同的網絡模型根據他們的後驗概率在整個網絡空間中跳動。為使得網絡模型能更好地在不同模型中轉換，我們把局部小網絡根據他們的可能性分配給他們相應的概率值。這些可能的局部小網絡是在數據前期處理中通過卡方檢驗得到的。和其他同類方法一樣， 雖然我們的方法也同樣面臨著在一個很大的網絡空間中搜索的難題， 但我們的方法能達到一個更高的推斷精度。同時，我們的方法所對應的計算量也是在可接收範圍之內。 / In bioinformatics, the gene regulatory network inference is gaining intensive attention nowadays. Various network models have been used to describe gene regulatory relationships, including deterministic Boolean networks, probabilistic Boolean networks, Bayesian networks, etc. This dissertation is focused on data-based Boolean network reconstruction. Many methods have been proposed to infer this discrete network structure. For example, the REVEAL algorithm and the Best-Fit Extension method are popular and perform well for the networks with limited total number of nodes. However, existing methods didn't take full consideration of the ubiquitous noise across the network and the structure uncertainty, which makes these algorithms unsatisfactory in real applications. In this dissertation, we use a full Bayesian approach to explore the space of probabilistic Boolean networks. To compare the relative fitness of networks to the input data, we design novel Markov chain Monte Carlo algorithms to jump among con rained networks according to the joint posterior probability. To facilitate the transdimensional move, high proposing probabilities are assigned to more likely subnetwork models as judged by chi-square tests in the preprocessing step. Although faced with the same difficulty of searching in a huge structure space as other methods, our algorithm is expected to reconstruct the Boolean network in a more accurate and comprehensive manner with a bearable computing cost. / Detailed summary in vernacular field only. / Han, Shengtong. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2012. / Includes bibliographical references (leaves 94-105). / Abstract also in Chinese. / Chapter 1 --- Introduction --- p.1 / Chapter 2 --- Technical Background --- p.5 / Chapter 2.1 --- Classical Boolean Network --- p.5 / Chapter 2.1.1 --- Definition --- p.5 / Chapter 2.1.2 --- Dynamic Properties --- p.8 / Chapter 2.2 --- Probabilistic Boolean Network --- p.9 / Chapter 2.2.1 --- Definition --- p.9 / Chapter 2.2.2 --- Dynamic Properties --- p.11 / Chapter 3 --- Bayesian Framework for Boolean Network Modeling --- p.12 / Chapter 3.1 --- Introduction --- p.12 / Chapter 3.2 --- Network Modeling --- p.15 / Chapter 3.2.1 --- Subnetwork Modeling --- p.15 / Chapter 3.2.2 --- Full Network Modeling --- p.21 / Chapter 3.2.3 --- Prior & Posterior Distributions --- p.23 / Chapter 4 --- Network Inference-MCMC --- p.29 / Chapter 4.1 --- Introduction --- p.29 / Chapter 4.2 --- Proposal Subnetwork Construction --- p.30 / Chapter 4.3 --- Network Structure Updating --- p.33 / Chapter 4.3.1 --- Individual Network Updating Moves --- p.33 / Chapter 4.3.2 --- Overall Network Updating Procedure --- p.37 / Chapter 4.3.3 --- The Core Metroplis-Hasting Algorithm --- p.37 / Chapter 4.4 --- Convergence Diagnostic --- p.40 / Chapter 4.5 --- Model Selection --- p.41 / Chapter 4.5.1 --- AIC, BIC --- p.42 / Chapter 4.5.2 --- Bayes Factor --- p.42 / Chapter 4.5.3 --- Reversible Jump MCMC --- p.43 / Chapter 4.5.4 --- Bayesian Model Averaging --- p.45 / Chapter 4.6 --- Computational Consideration --- p.46 / Chapter 5 --- Numerical Studies --- p.49 / Chapter 5.1 --- Simulation Studies --- p.49 / Chapter 5.1.1 --- Simulation for Synthetic Network Models with Small Number of Nodes --- p.50 / Chapter 5.1.2 --- Simulation for Synthetic Network Models with Large Number of Nodes --- p.64 / Chapter 5.2 --- Comparison with Other Methods --- p.68 / Chapter 5.2.1 --- Comparison Results --- p.71 / Chapter 5.2.2 --- Discussion --- p.72 / Chapter 6 --- Real Data Analysis --- p.74 / Chapter 6.1 --- A Real Cell Cycle Network --- p.74 / Chapter 6.2 --- Inference Result --- p.76 / Chapter 6.3 --- Discussion --- p.79 / Chapter 7 --- Summary and Discussion --- p.80 / Bibliography --- p.83 / Chapter A --- Data Pre-processing --- p.83 / Chapter A.1 --- Data Discretization --- p.83 / Chapter B --- Truth Tables for Commonly Used Basic Logic Functions --- p.85 / Chapter C --- All Distribution Tables for Gene Pairs and Gene Triplets --- p.86 / Chapter C.1 --- Distribution Assumptions for Input Gene Pairs --- p.86 / Chapter C.2 --- Distribution Assumptions for Gene Triplets --- p.87 / Chapter D --- Pseudo Code of the Algorithm --- p.91 / Chapter D.1 --- Case 1: In-degree=1 --- p.91 / Chapter D.2 --- Case 2: In-degree=2 --- p.93 / Chapter D.3 --- Case 3: In-degree=0 --- p.93

Han Wei. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2003. / Includes bibliographical references (leaves 60-61). / Abstracts in English and Chinese. / Abstract --- p.i / 摘要 --- p.ii / Acknowledgements --- p.iii / Contents --- p.iv / List of Figures --- p.vi / List of Tables --- p.vii / Chapter Chapter 1 --- Introduction --- p.1 / Chapter 1.1. --- Speech Recognition --- p.1 / Chapter 1.2. --- ASIC Design with HDLs --- p.3 / Chapter Chapter 2 --- Theory of HMM-Based Speech Recognition --- p.6 / Chapter 2.1. --- Speaker-Dependent and Speaker-Independent --- p.6 / Chapter 2.2. --- Frame and Feature Vector --- p.6 / Chapter 2.3. --- Hidden Markov Model --- p.7 / Chapter 2.3.1. --- Markov Model --- p.8 / Chapter 2.3.2. --- Hidden Markov Model --- p.9 / Chapter 2.3.3. --- Elements of an HMM --- p.10 / Chapter 2.3.4. --- Types of HMMs --- p.11 / Chapter 2.3.5. --- Continuous Observation Densities in HMMs --- p.13 / Chapter 2.3.6. --- Three Basic Problems for HMMs --- p.15 / Chapter 2.4. --- Probability Evaluation --- p.16 / Chapter 2.4.1. --- The Viterbi Algorithm --- p.17 / Chapter 2.4.2. --- Alternative Viterbi Implementation --- p.19 / Chapter Chapter 3 --- HMM-based Isolated Word Recognizer Design Methodology …… --- p.20 / Chapter 3.1. --- Speech Recognition Based On Single Mixture --- p.23 / Chapter 3.2. --- Speech Recognition Based On Double Mixtures --- p.25 / Chapter Chapter 4 --- VLSI Implementation of the Speech Recognizer --- p.29 / Chapter 4.1. --- The System Requirements --- p.29 / Chapter 4.2. --- Implementation of a Speech Recognizer with a Single-Mixture HMM --- p.30 / Chapter 4.3. --- Implementation of a Speech Recognizer with a Double-Mixture HMM --- p.39 / Chapter 4.4. --- Extend Usage in High Order Mixtures HMM --- p.46 / Chapter 4.5. --- Pipelining and the System Timing --- p.50 / Chapter Chapter 5 --- Simulation and IC Testing --- p.53 / Chapter 5.1. --- Simulation Result --- p.53 / Chapter 5.2. --- Testing --- p.55 / Chapter Chapter 6 --- Discussion and Conclusion --- p.58 / Reference --- p.60 / Appendix I Verilog Code of the Double-Mixture HMM Based Speech Recognition IC (RTL Level) --- p.62 / Subtracter --- p.62 / Multiplier --- p.63 / Core_Adder --- p.65 / Register for X --- p.66 / Subtractor and Comparator --- p.67 / Shifter --- p.68 / Look-Up Table --- p.71 / Register for Constants --- p.79 / Register for Scores --- p.80 / Final Score Register --- p.84 / Controller --- p.86 / Top --- p.97 / Appendix II Chip Microphotograph --- p.103 / Appendix III Pin Assignment of the Speech Recognition IC --- p.104 / Appendix IV The Testing Board of the IC --- p.108

Chung Po-Yi. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2003. / Includes bibliographical references (leaves 33-34). / Abstracts in English and Chinese. / Chapter 1 --- Introduction --- p.1 / Chapter 2 --- Bayesian Approach --- p.4 / Chapter 2.1 --- Model Description --- p.5 / Chapter 2.2 --- Identification --- p.6 / Chapter 2.3 --- Bayesian Analysis of the Model --- p.8 / Chapter 2.3.1 --- Posterior Analysis --- p.8 / Chapter 2.3.2 --- The Gibbs Sampler --- p.9 / Chapter 2.3.3 --- Conditional Distributions --- p.10 / Chapter 2.4 --- Bayesian Estimation --- p.13 / Chapter 3 --- Two-stage Partition Approach --- p.15 / Chapter 3.1 --- First Stage: PRELIS --- p.15 / Chapter 3.2 --- Second Stage: LISREL --- p.17 / Chapter 3.2.1 --- Model Description --- p.17 / Chapter 3.2.2 --- Identification --- p.17 / Chapter 3.2.3 --- LISREL Analysis of the Model --- p.18 / Chapter 4 --- Comparison --- p.19 / Chapter 4.1 --- Simulation Studies --- p.19 / Chapter 4.2 --- Real Data Studies --- p.28 / Chapter 5 --- Conclusion & Discussion --- p.30 / Chapter A --- Tables for the Two Approaches --- p.35 / Chapter B --- Manifest variables in the ICPSR examples --- p.51 / Chapter C --- PRELIS & LISREL Scripts for Simulation Studies --- p.52

Networks arise in nearly every branch of science, from biology and physics to sociology and economics. A signature of many network datasets is strong local dependence, which gives rise to phenomena such as sparsity, power law degree distributions, clustering, and structural heterogeneity. Statistical models of networks require a careful balance of flexibility to faithfully capture that dependence, and simplicity, to make analysis and inference tractable. In this dissertation, we introduce a class of models that insert one network edge at a time via a random walk, permitting the location of new edges to depend explicitly on the structure of the existing network, while remaining probabilistically and computationally tractable. Connections to graph kernels are made through the probability generating function of the random walk length distribution. The limiting degree distribution is shown to exhibit power law behavior, and the properties of the limiting degree sequence are studied analytically with martingale methods. In the second part of the dissertation, we develop a class of particle Markov chain Monte Carlo algorithms to perform inference for a large class of sequential random graph models, even when the observation consists only of a single graph. Using these methods, we derive a particle Gibbs sampler for random walk models. Fit to synthetic data, the sampler accurately recovers the model parameters; fit to real data, the model offers insight into the typical length scale of dependence in the network, and provides a new measure of vertex centrality. The arrival times of new vertices are the key to obtaining results for both theory and inference. In the third part, we undertake a careful study of the relationship between the arrival times, sparsity, and heavy tailed degree distributions in preferential attachment-type models of partitions and graphs. A number of constructive representations of the limiting degrees are obtained, and connections are made to exchangeable Gibbs partitions as well as to recent results on the limiting degrees of preferential attachment graphs.

Novel seasonal hidden Markov models (SHMMs) for stochastic time series with periodically varying characteristics are developed. Nonlinear interactions among SHMM parameters prevent the use of the forward-backward algorithms which are usually used to fit hidden Markov models to a data sequence. Instead, Powell's direction set method for optimizing a function is repeatedly applied to adjust SHMM parameters to fit a data sequence. SHMMs are applied to a set of meteorological data consisting of 9 years of daily rain gauge readings from four sites. The fitted models capture both the annual patterns and the short term persistence of rainfall patterns across the four sites.

In this thesis, we consider a robust state estimation problem for discrete-time, homogeneous, first-order, finite-state finite-alphabet hidden Markov models (HMMs). Based on Kolmogorov's Theorem on the existence of a process, we first present the Kolmogorov model for the HMMs under consideration. A new change of measure is introduced. The statistical properties of the Kolmogorov representation of an HMM are discussed on the canonical probability space. A special Kolmogorov measure is constructed. Meanwhile, the ergodicity of two expanded Markov chains is investigated. In order to describe the uncertainty of HMMs, we study probability distance problems based on the Kolmogorov model of HMMs. Using a change of measure technique, the relative entropy and the relative entropy rate as probability distances between HMMs, are given in terms of the HMM parameters. Also, we obtain a new expression for a probability distance considered in the existing literature such that we can use an information state method to calculate it. Furthermore, we introduce regular conditional relative entropy as an a posteriori probability distance to measure the discrepancy between HMMs when a realized observation sequence is given. A representation of the regular conditional relative entropy is derived based on the Radon-Nikodym derivative. Then a recursion for the regular conditional relative entropy is obtained using an information state method. Meanwhile, the well-known duality relationship between free energy and relative entropy is extended to the case of regular conditional relative entropy given a sub-[special character]-algebra. Finally, regular conditional relative entropy constraints are defined based on the study of the probability distance problem. Using a Lagrange multiplier technique and the duality relationship for regular conditional relative entropy, a finite horizon robust state estimator for HMMs with regular conditional relative entropy constraints is derived. A complete characterization of the solution to the robust state estimation problem is also presented.

Intelligent agents are designed to interact with, and learn about, their environment so that they can act purposefully towards a goal. One class of problems encountered in building such agents is learning how to respond to dynamic systems with a continuous state space. The goals of this dissertation are to develop a framework for understanding the behaviour of partitioned dynamic systems with continuous underlying state and to translate this framework into algorithms which adaptively form a partition of the continuous space such that the partitioned system is more easily learned and controlled, and such that the control law may be easily explained in intuitive ways. Currently, algorithms which learn a control policy for partitioned continuous state space systems treat the partitioned system as an approximation to a Markov chain. I give conditions for the partitioned system to be a Markov chain, a semi-Markov process and a new class of system, a weak-semi-Markov process. The weak-semi-Markov model is shown to model partitioned dynamic systems with greater economy than other surveyed models. The behaviour of a partitioned state space system in the area around the region boundaries is also considered. I use the theory of sliding surfaces, and some heuristic arguments to recommend region boundary shape and position. The concept of 'staying on the boundary' then becomes a robust and relatively easy subgoal within the control algorithm. The concept of 'reaching the sliding surface' as a subgoal is used as the basis for an intuitive explanation of the learnt controller. I present an algorithm based on this concept which explains the behaviour of a learnt controller in ways not previously available to a machine learning algorithms. Finally, the Markov Property and the theory of Sliding Mode Control are used as the basis of a class of recursive algorithms. These algorithms adaptively find a partition, and simultaneously use this partition in conjunction with one of five reinforcement learning algorithms to find a control policy based on that partition. This technique is shown to work very well in learning, controlling and explaining a variety of physical systems, from a monorail to a container crane.

在求無限期非均質馬可夫決策過程（nonhomogeneous Markov decisinon processes）第一期的的最佳解時，我們通常要將它表示成有限期的動態規劃問題。動態規劃可以用合成函數型式表示，也可以用最常見的線性規劃型式表示。 　　合成函數型式在傳統上是一直被認為「中看而不中用」，動態規劃的教科書中，只有在開場白中，介紹一下這種簡潔、漂亮的數學型式，然後就被完全打入冷宮，認為線性規劃型式才是真正實用、真正能讓電腦去執行求解的型式。在一般期刊的文獻中甚至根本不提這種表示法，而是花大篇篇幅在它所衍生的線性規劃技術上作文章，最典型的例子是Bean, Hopp and Duenyas(1992) 在OR期刊所發表的論文。 　　本文將完全針對這個問題的合成函數型式，討論它的一些性質，我們可以利用這些性質，設計出一個非常簡單、有效率的演算法。 / Hopp, Bean and Duenyas(1992) formulate a mixed integer program (MIP) to determine whether a finite time horizon is a forecast horizon in a nonhomogeneous Markov decision process(NMDP). Their formula are solved by complex Bender's decomposition In this thesis, we make an examination in details of the contraction property and affine mapping property of NMDP. By these properties we are relieved of the complex MIP formula and Bender's decomposition algorithm. The main contribution of the thesis is to show that it is not necessary to determine the optimal policies by running through the whole feasible solution space of their MIP problem. We only need to check a finite number of vertices at a polyhedral set shaped by the solution of the NMDP. The analysis shows insights into the NMDP and facilitate the prosess in determining the forecast horizon. Furthermore, this NMDP formulation is presented in the form of a simple dynamic function which is different from the linear program presented by Hopp, Bean and Duenyas.

Thesis (Ph. D.)--University of Missouri-Columbia, 2007. / The entire dissertation/thesis text is included in the research.pdf file; the official abstract appears in the short.pdf file (which also appears in the research.pdf); a non-technical general description, or public abstract, appears in the public.pdf file. Title from title screen of research.pdf file (viewed on December 18, 2007) Vita. Includes bibliographical references.