PermutMatrix is a work space designed to graphically explore gene expression data. It relies on the graphical approach introduced by Eisen and also offers several methods for the optimal reorganization of rows and columns of a numerical dataset. For example, several methods are proposed for optimal reorganization of the leaves of a hierarchical clustering tree, along with several seriation or unidimensional scaling methods that do not require any preliminary hierarchical clustering. This program, developed for MS Windows, with MS-Visual C++, has a clear and efficient graphical interface. Large datasets can be thoroughly and quickly analyzed.

We present a new model, derived from the hidden Markov model (HMM), to learn Boolean vector sequences. Our HMM with patterns (HMMP) is a simple, hybrid, and interpretable model that uses Boolean patterns to define emission probability distributions attached to states. Vectors consistent with a given pattern are equally probable, while inconsistent ones have probability zero to be emitted. We define an efficient learning algorithm for this model, which relies on the maximum likelihood principle, and proceeds by iteratively simplifying the structure and updating the parameters of an initial specific HMMP that represents the learning sequences. HMMPs and our learning algorithm are applied to the built-in self-test (BIST) for integrated circuits, which is one of the key microelectronic problems. An HMMP is learned from a test sequence set that covers most of the potential faults of the circuit at hand. Then, this HMMP is used as test sequence generator. The experiments carried out show that learned HMMPs have a very high fault coverage.

We propose a novel BIST technique for non-scan sequential circuits which does not modify the circuit under test. It uses a learning algorithm to build a hardware test sequence generator capable of reproducing the essential features of a set of precomputed deterministic test sequences. We use for this purpose two new models called hidden Markov model with patterns and independence model with patterns. Compared to existing methods, the proposed technique exhibits a very high fault coverage, including performance testing, at the expense of a low silicon area overhead.

Supervised classification has already been the subject of numerous studies in the fields of Statistics, Pattern Recognition and Artificial Intelligence under various appellations which include discriminant analysis, discrimination and concept learning. Many practical applications relating to this field have been developed. New methods have appeared in recent years, due to developments concerning Neural Networks and Machine Learning. These "hybrid" approaches share one common factor in that they combine symbolic and numerical aspects. The former are characterized by the representation of knowledge, the latter by the introduction of frequencies and probabilistic criteria. In the present study, we shall present a certain number of hybrid methods, conceived (or improved) by members of the SYMENU research group. These methods issue mainly from Machine Learning and from research on Classification Trees done in Statistics, and they may also be qualified as "rule-based". They shall be compared with other more classical approaches. This comparison will be based on a detailed description of each of the twelve methods envisaged, and on the results obtained concerning the "Waveform Recognition Problem" proposed by Breiman et al.,4 which is difficult for rule based approaches.