. Research at the University of Geneva reflects one of the main trends in machine learning today---the trend towards multistrategy learning. After attempts to compare a number of symbolic and connectionist learning strategies in the MELANIE project, increased involvement in neural network research led to the construction of FunNet, an environment for the specification and simulation of modular systems in which heterogeneous neural models and learning methods are combined in accordance with established theoretical constraints. FunNet is now used as the neural kernel of a software platform for integrating symbolic and connectionist processing. One architecture developed in this platform extends multistrategy learning to cover these two major AI paradigms: it uses symbolic rule learning as a metalevel tool for improving connectionist generalization. KEYWORDS. Machine learning, Hybrid learning, Neural Networks, Symbolic/connectionist integration, Symbolic/connectionist hybrid sytems 1. Int...

This work proposes an original approach to musical score recognition, a particular case of high-level document analysis. In order to overcome the limitations of existing systems, we propose an architecture which allows for a continuous and bidirectional interaction between high-level knowledge and low-level data, and which is able to improve itself over time. This architecture is made of three cooperating layers, one made of parametrized feature detectors, another working as an object-oriented knowledge repository and the other as a supervising metaprocessor. The key of this architecture is that all relations between pairs of domain concepts are represented in terms of measurable features of the input score. This makes relations learnable. We provide concrete examples, and show how this architecture is adequate for modelling and processing knowledge. Keywords: Musical score recognition Knowledge-based recognition Automatic learning and adaptation Neural networks 1 Introduction This...