Classes taught by CSLU Faculty
CSE540 Neural Network Algorithms and Architectures 3 credits A few years ago neural networks were just an emerging technology. Today they comprise a mature set of tools for nonlinear statistical modeling, yet remain an area of active research with new paradigms and applications. Neural nets are deployed commercially to applications in pattern recognition, prediction, control, and data mining. If you need a grasp of this technology for engineering applications, or simply want a sound theoretical and practical introduction, this course will get you there. Description: This course introduces the fundamentals of connectionist and neural network models. Paradigms for both unsupervised and supervised learning are covered. Topics include introduction to neural processing elements, Hebbian learning, LMS and back propagation algorithms, competitive learning, computational capability, and elements of statistical pattern recognition. Specific architectures covered include Hopfield nets, single and multilayer Perceptrons, and Kohonen maps. Programming projects involve network simulations and application problems. CSE546 Data and Signal Compression 3 credits Description: The need for signal and data compression is ubiquitous in image, video, and speech processing, finance, and computational science. Where data stores become very large (e.g. video, finance, earth science), the need is not met by simple lossless file compression schemes, and we must turn to sophisticated coding techniques. This course addresses both the theoretical basis and practical algorithms for data and signal compression. Topics include loss-less entropy based coding including Huffman and Lempel-Ziv, and lossy compression techniques including: scalar quantizers, transform coding (Karhunen-Loeve, DCT, and nonlinear transform codes), predictive coding, vector quantization, adaptive codes, and wavelets. The relation between compression schemes and probabilistic data modeling is emphasized in conjunction with each technique. Application to speech, image, and video coding are discussed. Students will have the opportunity to design compression schemes for such diverse applications as earth science data, finance, speech, or video depending on their specific interests. Text: Vector Quantization and Signal Compression, Gersho and Gray, Kluwer Academic. CSE547 Statistical Pattern Recognition 3 credits Speech recognition, image understanding, event detection, advanced signal processing and data mining all use methods and conceptual paradigms in statistical pattern recognition. Though pattern recognition technologies continue to evolve, all techniques and applications call for a thorough knowledge of the theoretical and practical cornerstones. By the end of this course, you'll have a thorough, working knowledge of the basics of statistical modeling, estimation, and performance evaluation common to all existing technologies; the principles that provide a framework for the understanding and evaluation of new technologies; and new problem domains. Description: Theory and practice of statistical pattern recognition. Students will learn fundamental theory and practices that are common to a broad range of pattern recognition applications and technologies, and apply principles to real-world examples. The emphasis is on developing tools, both theoretical and practical, that provide grounding in pattern recognition problems and methods; rather than on showcasing particular technologies. The course will benefit those whose work may use any of a variety of recognition technologies in broad-ranging applications such as speech and image processing, data mining, finance. Topics include: random vectors, detection problems (binary decision problems), likelihood ratio tests, ROC curves, parametric and non-parametric density estimation, classification models, theoretical error bounds and practical error estimation through cross-validation. Maximum likelihood and Bayesian parameter estimation. Feature extraction for dimensionality reduction, and for classification. CSE550 Spoken Language Systems 3 credits Spoken language systems will revolutionize how people interact with machines, replacing the keyboard and mouse with natural conversations. These systems will act like helpful human assistants and teachers for information access, commercial transactions, and learning. You'll review the state of the art in speech technologies, including speech recognition and understanding, speech synthesis and facial animation, and provides hands-on experience developing real world systems with CSLU's state of the art authoring tools. Description: In the not too distant future, spoken language systems will revolutionize human-computer interaction by enabling natural conversations between people and machines. In addition to telephony applications, such as voice browsing of the web, these systems also will support face-to-face communication with intelligent animated agents. These animated human-like agents will combine acoustic information with the speaker's facial cues and gestures to understand speech, and produce natural and expressive speech with accurate facial movements and expressions. This course reviews the state of the art in human language technology, and explains how key technologies are combined to produce spoken language systems. The course combines lectures by experts in the field with hands-on experience using and building spoken language systems using the CSLU Toolkit. The course materials are included in http://www.cse.ogi.edu/CSLU/HLTsurvey/HLTsurvey.html. CSE551 Structure of Spoken Language 3 credits The future of human-computer communication increasingly will use speech. Creating machines that interact with people using speech requires a good understanding of the acoustic and symbolic structure of language and of the capabilities and limitations of current spoken language systems. In CSE551, you'll examine what we know about speech and how this knowledge is used in speech recognition and synthesis. Description: This course provides a foundation for subsequent learning and research in computer speech recognition. We examine the structure of spoken English through selected readings in speech perception and acoustic phonetics and examination of visual displays of speech. The goals are to understand the acoustic cues for each major phonetic category, understand how these cues are affected by context, understand the perceptual strategies that listeners use to understand speech, and evaluate the assumption that speech can be described as an ordered sequence of phonetic segments CSE552 Hidden Markov Models for Speech Recognition 3 credits Description: Hidden Markov Model-based technology is used widely in today's speech recognition systems. This course is an introduction to speech recognition using HMM technology. Topics include the theory of Hidden Markov Models (discrete, semi-continuous, and continuous) and their applications to speech recognition, along with the basic mathematics (probability theory, statistics, stochastic process, information theory, and signal processing) that are necessary for speech recognition. The course is focused on understanding the theory behind these fundamental technologies, and applying the technology to develop speech recognition systems. CSE560 Artificial Intelligence 3 credits This course introduces you to declaratively representing information using rich knowledge representation schemes with formal semantics. You?ll learn to reason about this information in order to draw new conclusions, make consistent assumptions, or plan new actions. This reasoning process is at the heart of building intelligent agent-based systems. The theory in this course is balanced by building working programs in the logic-based programming language Prolog. Description: This course surveys the foundations and applications of symbolic approaches to artificial intelligence. The approach emphasizes the formal basis of automated reasoning and includes an introduction to programming in Prolog. Fundamentals covered include search, knowledge representation, automated inference, planning, nonmonotonic reasoning, and reasoning about belief. Applications include expert systems, natural language processing and agent architectures. ECE525 Analytical Techniques in Statistical Signal Processing and Communications 4 credits Development of the mathematical techniques needed to analyze systems involving random variables and/or stochastic processes with particular application to communications and instrumentation. Topics include Bayes Theorem (discrete and continuous forms), Tchebycheff inequality, Chernoff Bound, Central Limit Theorem, stationary processes and linear systems, mean square estimation, Poisson process, Gaussian process, Markoff process, and series representations. MATLAB and the MATLAB Statistics Tool Box are used in this course. ECE540 Auditory and Visual Processing by Human and Machine 4 credits Interaction between humans and machines could be greatly enhanced by machines that could communicate using human sensory signals such as speech and gestures. Knowledge of human information processing including audition, vision, and their combination is, therefore, critical in the design of effective human-machine interfaces. The course introduces selected phenomena in auditory and visual perception, and motor control. Students learn how to interpret empirical data, how to incorporate these data in models, and how to apply these models to engineering problems. The anthropomorphic (human-like) signal processing approach is illustrated on engineering models of perceptual phenomena. ECE541 Speech Processing 4 credits Speech is one of the most important means of communication. This course teaches theory of human speech production, properties of speech signal and techniques for its processing in speech coding, and automatic speech and speaker recognition. Emphasis is on active research in auditory modeling that exploits special properties of speech to improve performance of speech technology in practical applications. Prerequisites: ECE 540, ECE 551, or consent of instructor. ECE544 Introduction to Signals, Systems and Information Processing 4 credits This course provides the essential mathematical tools and analytical techniques needed for the analysis of continuous-time and discrete-time systems. Basic signal and system characteristics -- linearity, time-invariance, convolution and correlation -- are first examined from the time domain perspective. We then proceed to study a family of Transforms - Fourier Series, Fourier Integral Transform, Laplace Transform , Discrete Time Fourier Transform (DTFT) , Discrete Fourier Transform (DFT) and z-Transform -- which take the study of these systems to a deeper level and introduce a host of useful properties which the time perspective alone does not reveal. Basic applications taken from the areas of information processing, communication and control will serve to fill out the mathematically derived results. A greater portion of the syllabus in ECE 544 is allotted to continuous time signals/systems than to discrete time signals/systems, for reason that the latter are taken up in detail in other information processing courses, particularly ECE 551. A goal of the presentation in ECE 544 is to impart the essential unity of all the Transforms and the almost perfect correspondence of approach in continuous-time and discrete-time contexts. You then become a well equipped practitioner who knows the way around the entire territory. This course is a useful prerequisite or corequisite to ECE 551 and all other courses in the information processing area. ECE551 Introduction to Digital Signal Processing 4 credits Representation and analysis of discrete time signals and systems. Z-Transform, Discrete-Time Fourier Transform, and Discrete Fourier Transforms. Applications of the Fast Fourier Transform to high-speed computation of convolution and correlation products. Signal flow graph realizations of finite word-length implemented discrete time linear systems. Sampling and windowing techniques pertaining to the discrete time processing of continuous time signals. Analysis and design of recursive and nonrecursive digital filters. |