Analysis of sensory information using artificial neural networks and fuzzy logic

Feeding decisions are made by the insect central nervous system based on neural input from chemoreceptors responding to phytochemical stimuli. A computational model is being developed that simulates feeding decisions by a caterpillar choosing among acceptable and unacceptable plants. Inputs to the model are electrophysiological recordings from the gustatory neurons of the caterpillar. The objectives are to determine which components of the sensory data are important in the decision process, and to elucidate the rules by which these decisions are made. The three-stage model employs: (1) artificial neural networks (ANN) for processing and classifying action potentials; (2) an adaptive algorithm for integrating data from multiple sensory organs; and (3) fuzzy logic for determining the likelihood of feeding. This third “decision module” is “trained” to associate chemosensory data with behavioral responses to the same stimulus. When trained with response to a variety of acceptable and unacceptable plants, the model can be challenged with sensory responses to a untested stimulus and asked to predict the feeding decision. These simulated decisions can then be compared with experimentally derived behavioral data to verify the accuracy of the model. Analysis of the structure (e.g., the “synaptic weights”) of a successfully trained ANN provides information about which activities and interactions of the chemosensory neurons shape the decision-making process