A probabilistic network is a graphical model that encodes probabilistic relationships between variables of interest. Such a model records qualitative influences between variables in addition to the numerical parameters of the probability distribution. As such it provides an ideal form for combining prior knowledge, which might be limited solely to experience of the influences between some of the variables of interest, and data. In this paper, we first show how data can be used to revise initial estimates of the parameters of a model. We then progress to showing how the structure of the model can be revised as data is obtained. Techniques for learning with incomplete data are also covered.

In this paper a theoretical framework for Bayesian adaptive learning of discrete HMM and semi-continuous one with Gaussian mixture state observation densities is presented. Corresponding to the well-known Baum-Welch and segmental k-means algorithms respectively for HMM training, formulations of MAP (maximum aposteriori) and segmental MAP estimation of HMM parameters are developed. Furthermore, a computationally efficient method of the segmental quasi-Bayes estimation for semi-continuous HMM is also presented. The important issue of prior density estimation is discussed and a simplified method of moment estimate is given. The method proposed in this paper will be applicable to some problems in HMM training for speech recognition such as sequential or batch training, model adaptation, and parameter smoothing, etc.

Automated visual-tracking of cell populations in vitro using time-lapse phase contrast microscopy enables quantitative, systematic and high-throughput measurements of cell behaviors. These measurements include the spatiotemporal quantification of cell migration, mitosis, apoptosis, and the reconstruction of cell lineages. The combination of low signal-to-noise ratio of phase contrast microscopy images, high and varying densities of the cell cultures, topological complexities of cell shapes, and wide range of cell behaviors poses many challenges to existing tracking techniques. This paper presents a fully-automated multi-target tracking system that can efficiently cope with these challenges while simultaneously tracking and analyzing thousands of cells observed using time-lapse phase contrast microscopy. The system combines bottom-up and top-down image analysis by integrating multiple collaborative modules, which exploit a fast geometric active contour tracker in conjunction with adaptive interacting multiple models (IMM) motion filtering and spatiotemporal trajectory optimization. The system, which was tested using a variety of cell populations, achieved tracking accuracy in the range of 86.9%-92.5%.

Recent advances in automatic speech recognition are accomplished by designing a plug-in maximum a posteriori decision rule such that the forms of the acoustic and language model distributions are specified and the parameters of the assumed distributions are estimated from a collection of speech and language training corpora. Maximum-likelihood point estimation is by far the most prevailing training method. However, due to the problems of unknown speech distributions, sparse training data, high spectral and temporal variabilities in speech, and possible mismatch between training and testing conditions, a dynamic training strategy is needed. To cope with the changing speakers and speaking conditions in real operational conditions for high-performance speech recognition, such paradigms incorporate a small amount of speaker and environment specific adaptation data into the training process. Bayesian adaptive learning is an optimal way to combine

Approximate nonparametric Bayes estimates calculated under a Dirichlet process prior are readily obtained in a wide range of models using a simple recursive algorithm. This chapter develops the recursion using elementary facts about nonparametric predictive distributions, and applies it to an interval censoring problem and to a Markov chain mixture model. S-Plus code is provided. 1 Introduction Sampling models that enforce relatively weak assumptions are naturally favored in many applications, but it is well known that the corresponding posterior computations can become very intensive when a Dirichlet process encodes prior uncertainty in the weakly specified part of the model. In all but the most simple models, posterior calculations involve a mixture of Dirichlet processes. As evidenced by companion chapters, advances in Markov chain Monte Carlo (MCMC) provide critical methodology for enabling these calculations, and have opened up a wide range of interesting applications to Dirichle...

We accept this thesis as conforming

We study the problem of learning to recognise objects in the context of autonomous agents.

The Temporal Coding Hypothesis of Miller and colleagues [7] suggests that animals integrate related temporal patterns of stimuli into single memory representations. We formalize this concept using quasi-Bayes estimation to update the parameters of a constrained hidden Markov model. This approach allows us to account for some surprising temporal e#ects in the second order conditioning experiments of Miller et al. [1, 2, 3], which other models are unable to explain.

A perceptual system encompasses an important perspective of the interface between human and an interacting system. In this context, an approach is presented to establish a natural and effective interaction between an operator and a measurement system. The system output is acting as a social agent, based on sensory data and is presenting impressions as facial expressions as well as conclusions in form of natural language based identification. The proposed method has been evaluated in an experimental perceptual odour system set up using an artificial olfactory system.

We present an optimization model to find joint inspection and maintenance policies for infrastructure facilities under performance model uncertainty. The objective in the formulation is to minimize the total expected social cost of managing facilities over a finite planning horizon. As in recent optimization models, performance model uncertainty is accounted for by representing facility deterioration as a mixture of known models taken from a finite set. The mixture proportions are assumed to be continuous random variables, with probability densities that are updated over time. In this paper, we relax the assumptions of fixed and error-free inspections. We present a parametric study to analyze the effect of initial performance model uncertainty and bias on the expected total cost of managing a facility. The main observation is that reducing the initial variance in model uncertainty may be more important than reducing the initial bias. Our study also shows that cost savings can result from relaxing the constraint of a fixed inspection schedule.