ABSTRACT
A standard sales pitch for connectionism says that network models share elementary properties in common with the brain, such as massive parallelism, neuron-like computing units, synapse-like weights, fault tolerance, and adaptability [1, 2, 3]. Connectionists have been keen to stress these similarities, and extol connectionism as a potentially powerful tool for studying the nervous system. However, if connectionist models are to help us understand the brain, they must be applied in a manner that is consistent with the methods and goals of neuroscience. In particular, connectionist modeling techniques must be integrated with anatomical, physiological, and behavioral research to develop plausible theories of how the nervous system functions.
