Humans use visual as well as auditory speech signals to recognize spoken words. A variety of systems have been investigated for per(cid:173) forming this task. The main purpose of this research was to sys(cid:173) tematically compare the performance of a range of dynamic visual features on a speechreading task. We have found that normal(cid:173) ization of images to eliminate variation due to translation, scale, and planar rotation yielded substantial improvements in general(cid:173) ization performance regardless of the visual representation used. In addition, the dynamic information in the difference between suc(cid:173) cessive frames yielded better performance than optical-flow based approaches, and compression by local low-pass filtering worked sur(cid:173) prisingly better than global principal components analysis (PCA).

This paper describes how the early visual process of contour organ(cid:173) isation can be realised using the EM algorithm. The underlying computational representation is based on fine spline coverings. Ac(cid:173) cording to our EM approach the adjustment of spline parameters draws on an iterative weighted least-squares fitting process. The expectation step of our EM procedure computes the likelihood of the data using a mixture model defined over the set of spline cover(cid:173) ings. These splines are limited in their spatial extent using Gaus(cid:173) sian windowing functions.

We consider the problem of prediction of stationary time series, using the architecture known as mixtures of experts (MEM). Here we suggest a mixture which blends several autoregressive models. This study focuses on some theoretical foundations of the predic(cid:173) tion problem in this context. More precisely, it is demonstrated that this model is a universal approximator, with respect to learn(cid:173) ing the unknown prediction function . This statement is strength(cid:173) ened as upper bounds on the mean squared error are established.

A modification is described to the use of mean field approxima(cid:173) tions in the E step of EM algorithms for analysing data from latent structure models, as described by Ghahramani (1995), among oth(cid:173) ers. The modification involves second-order Taylor approximations to expectations computed in the E step. The potential benefits of the method are illustrated using very simple latent profile models.

We have explored two approaches to recogmzmg faces across changes in pose. First, we developed a representation of face images based on independent component analysis (ICA) and compared it to a principal component analysis (PCA) representation for face recognition. The ICA basis vectors for this data set were more spatially local than the PCA basis vectors and the ICA representa(cid:173) tion had greater invariance to changes in pose. Second, we present a model for the development of viewpoint invariant responses to faces from visual experience in a biological system. The temporal continuity of natural visual experience was incorporated into an attractor network model by Hebbian learning following a lowpass temporal filter on unit activities.

Unsupervised learning algorithms based on convex and conic en(cid:173) coders are proposed. The encoders find the closest convex or conic combination of basis vectors to the input. The learning algorithms produce basis vectors that minimize the reconstruction error of the encoders. The convex algorithm develops locally linear models of the input, while the conic algorithm discovers features. Both al(cid:173) gorithms are used to model handwritten digits and compared with vector quantization and principal component analysis.

The following investigates the use of single-neuron learning algo(cid:173) rithms to improve the performance of text-retrieval systems that accept natural-language queries. A retrieval process is explained that transforms the natural-language query into the query syntax of a real retrieval system: the initial query is expanded using statis(cid:173) tical and learning techniques and is then used for document ranking and binary classification. The results of experiments suggest that Kivinen and Warmuth's Exponentiated Gradient Descent learning algorithm works significantly better than previous approaches.

The inverse of the Fisher information matrix is used in the natu(cid:173) ral gradient descent algorithm to train single-layer and multi-layer perceptrons. We have discovered a new scheme to represent the Fisher information matrix of a stochastic multi-layer perceptron. Based on this scheme, we have designed an algorithm to compute the natural gradient. When the input dimension n is much larger than the number of hidden neurons, the complexity of this algo(cid:173) rithm is of order O(n). It is confirmed by simulations that the natural gradient descent learning rule is not only efficient but also robust.

We derive the correspondence between regularization operators used in Regularization Networks and Hilbert Schmidt Kernels appearing in Sup(cid:173) port Vector Machines. More specifica1ly, we prove that the Green's Func(cid:173) tions associated with regularization operators are suitable Support Vector Kernels with equivalent regularization properties. As a by-product we show that a large number of Radial Basis Functions namely condition(cid:173) ally positive definite functions may be used as Support Vector kernels.

This paper describes a new approach to extracting 3D perspective structure from 2D point-sets. The novel feature is to unify the tasks of estimating transformation geometry and identifying point(cid:173) correspondence matches. Unification is realised by constructing a mixture model over the bi-partite graph representing the correspon(cid:173) dence match and by effecting optimisation using the EM algorithm. According to our EM framework the probabilities of structural cor(cid:173) respondence gate contributions to the expected likelihood function used to estimate maximum likelihood perspective pose parameters. This provides a means of rejecting structural outliers.