Narrow Results

In practical applications, recognition accuracy is sometimes not the only criterion; capability to reject erroneous patterns might also be needed. We show that there is a trade-off between these two properties. An efficient solution to this trade-off is brought about by the use of different algorithms implemented in various modules, i.e. multi-modular architectures.

We present a general mechanism for designing and training multi-modular architectures, integrating various neural networks into a unique pattern recognition system, which is globally trained. It is possible to realize, within the system, feature extraction and recognition in successive modules which are cooperatively trained. We discuss various rejection criteria for neural networks and multi-modular architectures.

We then give two examples of such systems, study their rejection capabilities and show how to use them for segmentation. In handwritten optical character recognition, our system achieves performances at state-of-the-art level, but is eight times faster. In human face recognition, our system is intended to work in the real world.

First the paper explains why fuzzy inference system can be regarded as just another interesting grey-box way of approximating non-linear mapping. Then it contributes at clarifying the current confusion raised by a lot of works comparing or merging neural nets with fuzzy inference systems. Practical comparisons with RBF are performed which show that the small structural addition leading to fuzzy systems can be of interest for function identification. To face the curse of dimensionality problem, the paper presents an algorithm developed in a biological spirit and dedicated to the on-line incremental building of fuzzy systems for function approximation. It is called EFUSS (Evolving Fuzzy Systems Structure) and aims at automatically and incrementally finding the minimal number of membership functions along with their appropriate shaping. Its main characteristics are that the structural additions occur at a lower time scale than the parametric changes. They are guided by the endogenous dynamics of the parametric learning and aim at compensating for the weakest parts of the systems.