In this thesis, a two-word, three-word and four-word Mandarin phrases speech recognition system is studied and implemented. This system utilizes hidden Markov model, lexicon search strategy and tone recognition to select the initial phrase candidates and make the final decision. Experimental results indicate that using about one third of the total phrase population, 80%, 92% and 97% correct rates can be achieved for the 70,000 two-word, 24,000 three-word and 22,000 four-word phrases recognition problems respectively. Any spoken phrase can be found within 1 second, using a PC with Intel Celeron 2.4 GHz CPU and Red Hat Linux 9.0 operating system.

In this thesis, hidden Markov model, maximum likelihood ratio and lexicon search strategy are used to establish a Chinese resume inputting system. The resume contains five items: name introduction, gender, birth date, birth place and education. This system is developed using a PC with an Intel Pentium 1.6 GHz CPU and Red Hat Linux 9.0 operating system. For the speaker-dependent case, a resume can be completed within 45 seconds on the average.

The objective of this thesis is to design and implement a speech recognition system for one hundred thousand Chinese names. Mel frequency cepstrum coefficient, hidden Markov model and lexicon search strategy are utilized to choose the name candidates. Furthermore, a mandarin intonation technique is also incorporated into this system to increase the final speech recognition accuracy. The experimental results indicate that for the speaker dependent case, an 85% correct rate can be achieved by use of the proposed intonation classification scheme and the balanced monosyllable training database. The above correct rate has an increase of 8% over the previous method without using these two techniques. Under Redhat Linux 9.0 environment, a mandarin name can be recognized within 2 seconds by the use of a computer with Intel Celeron 2.4 GHz CPU.

Cloud computing provides business new working paradigm with the benefit of cost reduce and resource sharing. Tasks from different users may be performed on the same machine. Therefore, one primary security concern is whether user data is secure in cloud. On the other hand, hacker may facilitate cloud computing to launch larger range of attack, such as a request of port scan in cloud with virtual machines executing such malicious action. In addition, hacker may perform a sequence of attacks in order to compromise his target system in cloud, for example, evading an easy-to-exploit machine in a cloud and then using the previous compromised to attack the target. Such attack plan may be stealthy or inside the computing environment, so intrusion detection system or firewall has difficulty to identify it. The proposed detection system analyzes logs from cloud to extract the intensions of the actions recorded in logs. Stealthy reconnaissance actions are often neglected by administrator for the insignificant number of violations. Hidden Markov model is adopted to model the sequence of attack performed by hacker and such stealthy events in a long time frame will become significant in the state-aware model. The preliminary results show that the proposed system can identify such attack plans in the real network.

Wireless network analysis and simulations rely on accurate physical layer models. The increased interest in wireless network design and cross-layer design require an accurate and efficient physical layer model especially when a large number of nodes are to be studied and building the real network is not possible. For analysis of upper layer characteristics, a simplified physical layer model has to be chosen to model the physical layer. In this dissertation, the widely used two-state Markov model is examined and shown to be deficient for low to moderate signal-to-noise ratios. The physical layer statistics are investigated, and the run length distributions of the good and bad frames are demonstrated to be the key statistics for accurate physical layer modeling. A four-state Markov model is proposed for the flat Rayleigh fading channel by approximating the run length distributions with a mixture of exponential distributions. The transition probabilities in the four-state Markov model can be established analytically without having to run extensive physical layer simulations, which are required for the two-state Markov model. Physical layer good and bad run length distributions are compared and it is shown that the four-state Markov model reasonably approximates the run length distributions. Ns2 simulations are performed and the four-state Markov model provides a much more realistic approximation compared to the popular two-state Markov model. Achieving good results with the flat Rayleigh fading channel, the proposed four-state Markov model is applied to a few diversity channels. A coded orthogonal fre- quency division multiplexing (OFDM) system with a frequency selective channel and the Alamouti multiple-input multiple-output system are chosen to verify the accuracy of the four-state Markov model. The network simulation results show that the four-state Markov model approximates the physical layer with diversity channel well whereas the traditional two-state Markov model estimates the network throughput poorly. The success of adapting the four-state Markov model to the diversity channel also shows the flexibility of adapting the four-state Markov model to various channel conditions.

A Mandarin keyword spotting system based on LPC, VQ, discrete-time HMM and Viterbi algorithm is proposed in the thesis. Joining with a dialogue system, this keyword spotting platform is further refined to a prototype of Taiwan Railway Natural Language Reservation System. In the reservation process, five questions: name and ID number, departure station, destination station, train type and number of tickets, and time schedule are asked by the computer-dialogue attendant. Following by the customer¡¦s speech confirmation, electronic tickets can be correctly issued and printed within 90 seconds in a laboratory environment.

Existing methods for image-based location estimation generally attempt to recognize every photo independently, and their resulting reliance on strong visual feature matches makes them most suited for distinctive landmark scenes. We observe that when touring a city, people tend to follow common travel patterns---for example, a stroll down Wall Street might be followed by a ferry ride, then a visit to the Statue of Liberty or Ellis Island museum. We propose an approach that learns these trends directly from online image data, and then leverages them within a Hidden Markov Model to robustly estimate locations for novel sequences of tourist photos. We further devise a set-to-set matching-based likelihood that treats each ``burst" of photos from the same camera as a single observation, thereby better accommodating images that may not contain particularly distinctive scenes. Our experiments with two large datasets of major tourist cities clearly demonstrate the approach's advantages over traditional methods that recognize each photo individually, as well as a naive HMM baseline that lacks the proposed burst-based observation model. / text

This paper investigates optimal portfolio strategies in a market with partial information on the drift. The drift is modelled as a function of a continuous-time Markov chain with finitely many states which is not directly observable. Information on the drift is obtained from the observation of stock prices. Moreover, expert opinions in the form of signals at random discrete time points are included in the analysis. We derive the filtering equation for the return process and incorporate the filter into the state variables of the optimization problem. This problem is studied with dynamic programming methods. In particular, we propose a policy improvement method to obtain computable approximations of the optimal strategy. Numerical results are presented at the end. (author's abstract)

In today’s fast paced world, as the number of stations of television programming offered is increasing rapidly, time accessible to watch them remains same or decreasing. Sports videos are typically lengthy and they appeal to a massive crowd. Though sports video is lengthy, most of the viewer’s desire to watch specific segments of the video which are fascinating, like a home-run in a baseball or goal in soccer i.e., users prefer to watch highlights to save time. When associated to the entire span of the video, these segments form only a minor share. Hence these videos need to be summarized for effective presentation and data management. This thesis explores the ability to extract highlights automatically using MPEG-7 features and hidden Markov model (HMM), so that viewing time can be reduced. Video is first segmented into scene shots, in which the detection of the shot is the fundamental task. After the video is segmented into shots, extraction of key frames allows a suitable representation of the whole shot. Feature extraction is crucial processing step in the classification, video indexing and retrieval system. Frame features such as color, motion, texture, edges are extracted from the key frames. A baseball highlight contains certain types of scene shots and these shots follow a particular transition pattern. The shots are classified as close-up, out-field, base and audience. I first try to identify the type of the shot using low level features extracted from the key frames of each shot. For the identification of the highlight I use the hidden Markov model using the transition pattern of the shots in time domain. Experimental results suggest that with reasonable accuracy highlights can be extracted from the video.

With the development of sequencing technologies, more and more sequence variants are available for investigation. Different types of variants in the human genome have been identified, including single nucleotide polymorphisms (SNPs), short insertions and deletions (indels), and large structural variations such as large duplications and deletions. Of great research interest is the functional effects of these variants. Although many programs have been developed to predict the effect of SNPs, few can be used to predict the effect of indels or multiple variants, such as multiple SNPs, multiple indels, or a combination of both. Moreover, fine grained prediction of the functional outcome of variants is not available. To address these limitations, we developed a prediction framework, HMMvar, to predict the functional effects of coding variants (SNPs or indels), using profile hidden Markov models (HMMs). Based on HMMvar, we proposed HMMvar-multi to explore the joint effects of multiple variants in the same gene. For fine grained functional outcome prediction, we developed HMMvar-func to computationally define and predict four types of functional outcome of a variant: gain, loss, switch, and conservation of function. / Ph. D.