Particle Filtering

Many problems in signal processing can be modeled by a state-space model that evolves over time. As the system changes, observations become available. In a Bayesian sense, all our knowledge can be incorporated into the posterior distribution. In all but the simplest problems, this distribution is impossible to calculate analytically. Particle filtering is a method for approximating this posterior distribution based on sequential monte-carlo sampling techniques.

Independent Component Analysis

A central problem in the areas of signal processing, statistics, neural networks, etc. is to find suitable representations of multivariate data. Many techniques have been proposed to address this problem, e.g., principal component analysis, factor analysis, and projection pursuit, to name a few. Independent component analysis (ICA) is another such technique that has gained popularity in the past few years due to its robustness and wide applicability. ICA seeks a representation of the data in which the components have least statistical dependency and maximal non Gaussianity. In this talk I will review the framework for the application of ICA, viz., the underlying data model, the assumptions of ICA, and measures of non Gaussianity used in the analysis. Applications to medical data analysis, blind source separation, and blind deconvolution are also discussed.


In this presentation, we first give a short overview of what is biometric and what is going on in biometrics. Mapping a biometric authentication system to a pattern recognition system as the enrollment step can be treated as offline model and the authentication step can be treated as on-line model. So we can use some general techniques in pattern recognition to BAS. Review some classic biometric as fingerprint and iris for ideas of feature points extraction. And classify the system performance in several ways: basic system errors (FAR/FRR and roc), performance testing (matcher qualifying), biometric individuality (threshold selection and reprehensive samples selection) and system error revisited (confidence interval). Even introduce some advance topic as multi-biometric system as integration information, CMC (cumulative match curve) as another evaluation reference.

Quantum Information Theory

We begin with brief history and current status of quantum information theory. Hilbert space, Bloch sphere, Liouville space, ensemble of quantum states, and Von Neumann entropy are included as quantum theory postulates. Selected topics of quantum data compression, quantum information theory include Holevo bound and accessible information, quantum channel capacity, and also quantum entanglement and maximally entangled states. We sincerely thank Aikaterini D. Mandilara for her excellent and informative presentations.

Statistics of Natural Imagery

We study the statistical properties of images, image categories, and image decomposition. The applications and relative areas are also covered, with a focus on the analysis of human visual system, such as simple cell receptive field properties, sparse coding in V1, and neural representation of images.

Unsolved Problems in Information Theory

Two-way communication, distributed source coding, and achievable rates for pattern recognition systems are the three main subjects covered in spring 2005.

Data Embedding

Information theoretic analysis of information hiding and wavelet watermarking algorithms, especially those for JPEG2000, are covered.

Image Registration

Mutual information based methods, point-based methods, and surface-based methods for image registration are covered.

Transmit Beamforming for MIMO Radar Systems

A method to design cross correlation matrix of signals to achieve or approximate a desired spatial transmit beampattern is proposed. This is a possible direction of designing next generation radar systems which have complete flexibility in the choice of signals transmitted at each of their aperture.

The Concave-Convex Procedure

Theoretical foundations of CCCP and its relation with other iterative optimization methods, such as EM algorithm, variational bounding, Legendre Transforms and Sinkhorn¡¦s algorithm, are discussed.

Shape Theory

Stochastic Complexity

Information Theoretic Image Formation

Active Sensing and Active Testing

Communication Complexity

Random Point Processes

Content Based Image Retrieval

Overview 0000000 People 0000000 ITRRhome 0000000 Seminars