Biological Mechanisms for Learning: A Computational Model of Olfactory Learning in the Manduca sexta Moth, with Applications to Neural Nets.

The insect olfactory system, which includes the antennal lobe (AL), mushroombody (MB), and ancillary structures, is a relatively simple neural systemcapable of learning. Its structural features, which are widespread inbiological neural systems, process olfactory stimuli through a cascade ofnetworks where large dimension shifts occur from stage to stage and wheresparsity and randomness play a critical role in coding. Learning is partlyenabled by a neuromodulatory reward mechanism of octopamine stimulation of theAL, whose increased activity induces rewiring of the MB through Hebbianplasticity. Enforced sparsity in the MB focuses Hebbian growth on neurons thatare the most important for the representation of the learned odor. Based uponcurrent biophysical knowledge, we have constructed an end-to-end computationalmodel of the Manduca sexta moth olfactory system which includes the interactionof the AL and MB under octopamine stimulation. Our model is able to robustlylearn new odors, and our simulations of integrate-and-fire neurons match thestatistical features of in-vivo firing rate data. From a biologicalperspective, the model provides a valuable tool for examining the role ofneuromodulators, like octopamine, in learning, and gives insight into criticalinteractions between sparsity, Hebbian growth, and stimulation during learning.Our simulations also inform predictions about structural details of theolfactory system that are not currently well-characterized. From a machinelearning perspective, the model yields bio-inspired mechanisms that arepotentially useful in constructing neural nets for rapid learning from very fewsamples. These mechanisms include high-noise layers, sparse layers as noisefilters, and a biologically-plausible optimization method to train the networkbased on octopamine stimulation, sparse layers, and Hebbian growth.