Adaptive Synaptogenesis Constructs Neural Codes That Benefit Discrimination

Intelligent organisms face a variety of tasks requiring the acquisition of expertise within a specific domain, including the ability to discriminate between a large number of similar patterns. From an energy-efficiency perspective, effective discrimination requires a prudent allocation of neural resources with more frequent patterns and their variants being represented with greater precision. In this work, we demonstrate a biologically plausible means of constructing a single-layer neural network that adaptively (i.e., without supervision) meets this criterion. Specifically, the adaptive algorithm includes synaptogenesis, synaptic shedding, and bi-directional synaptic weight modification to produce a network with outputs (i.e. neural codes) that represent input patterns proportional to the frequency of related patterns. In addition to pattern frequency, the correlational structure of the input environment also affects allocation of neural resources. The combined synaptic modification mechanisms provide an explanation of neuron allocation in the case of self-taught experts.Author Summary: One neural correlate of being an expert is more brain volume—and presumably more neurons and more synapses—devoted to processing the input patterns falling within one's field of expertise. As the number of neurons in the neocortex does not increase during the learning period that begins with novice abilities and ends with expert performance, neurons must be viewed as a scarce resource whose connections are adjusted to be more responsive to inputs within the field of expertise and less responsive to input patterns outside this field. To accomplish this enhanced, but localized improvement of representational capacity, the usual theory of associative synaptic modification is extended to include synaptogenesis (formation of new synapses) and synaptic shedding (rejection of synapses by a postsynaptic neuron) in a manner compatible with the original, associative synaptic modification algorithm. Using some mathematically simplifying assumptions, a theory is developed that predicts the algorithm's eventual outcome on expert neuronal coding, and then without the simplifying assumptions, computational simulations confirm the theory’s predictions in long, but finite periods of simulation-time (i.e., finite-sampling leads to stable connections, and thus, stable expert encodings).