Murre (1992) takes the stance that psychological plausibility takes precedence over mathematical tractability, while not excluding the latter. Murre's volume is worth serious study, particularly the sections on psychological and biological plausibility.
2. This book, for the beginner with a little computing and mathematical competence, and an awareness of issues in cognitive psychology, is a better place to start then the obsolescent parallel distributed processing model which has been over-quoted. Hierarchical, layered, fully connected structures are not what the brain has, and backpropagation is not what it does, nor what it could be doing if it learns sequentially.
3. Murre builds from CALM modules of connected excitatory and inhibitory units, which have internal fixed structure; the intramodule connections are unmodifiable but the intermodule connections between excitatory units are modifiable. There are neurophysiological reasons why this structure is plausible, and it can exhibit both supervised and unsupervised learning, and interestingly can learn EXOR as well as form topological maps.
4. There are some affinities with ideas of Kohonen (1988), but CALM networks appear to be faster yet not improbabaly fast. Learning theorists will be interested in the stability of CALM networks in a learning mode and in the implementation of the distinction between explicit and implicit memory processes. One notes that parameter values have to be delicately adjusted to avoid what Murre calls reverberatory internal loops between modules; this point recalls Haken's (1987) insight that the brain must function just on the stable side of chaotic dynamics. The rather uncritical acceptance (p. 129) of Shepard's (1974) and Nosofksky's (1992) use of hypothetical metric spaces in psychology is unfortunate and not necessary to the general argument, indeed the biological features of these "spaces," Murre notes, are unsatisfactorily undefined.
5. For this reviewer, who has ready access to massive parallel computing, the technical appendices on programming were the part to read first. Neural network texts are necessarily ephemeral; ideas, hardware and software have an apparent half-life of about three years. This book in paperback is good value for the money, and worth serious study, particularly the sections on psychological and biological plausibility. Software to run on personal computers is said to be available.
Haken, H. (1987) (Ed.) Advanced Synergetics. Berlin: Springer-Verlag.
Kohonen, T. (1988) Representation of sensory information in self-organizing feature maps, and the relation of these maps to distributed memory networks. In: Cotterill, R. M. J. (Ed.) Computer Simulation in Brain Science. Cambridge: C.U.P., pp. 12-25.
Murre, J.M.J. (1992) Learning and Categorization in Modular Neural Networks. UK: Harvester/Wheatsheaf; US: Erlbaum
Murre, J.M.J. (1992) Precis of: Learning and Categorization in Modular Neural Networks. PSYCOLOQUY 3(68) categorization.1
Nosofksky, R. M. (1992) Similarity Scaling and Cognitive Process Models. Annual Review of Psychology, 43, 25-54.
Shepard, R. N. (1974) Representations of Structure in Similarity Data: Problems and Prospects. Psychometrika, 39, 373-421.