Insect cyborgs: Bio-mimetic feature generators improve machine learning accuracy on limited data

Machine learning (ML) classifiers always benefit from more informative input features. We seek to auto-generate stronger feature sets in order to address the difficulty that ML methods often experience given limited training data. A wide range of biological neural nets (BNNs) excel at fast learning, implying that they are adept at extracting informative features. We can thus look to BNNs for possible tools to improve performance of ML methods in this low-data regime. The insect olfactory network, though simple, learns new odors very rapidly, by means of three key elements: A competitive inhibition layer (Antennal Lobe, AL); a high-dimensional sparse plastic layer (Mushroom Body, MB); and Hebbian updates of synaptic weights. In this work, we deploy a computational model of the insect AL-MB as an automatic feature generator, attached as a front-end pre-processor so that its Readout Neurons provide new features, derived from the original features, for use by standard ML classifiers. We find that these ``insect cyborgs'', i.e. classifiers that are part-insect model and part-ML method, have significantly better performance than baseline ML methods alone on a vectorized MNIST dataset. Relative Test set accuracy improves by an average of 6% to 33% depending on baseline ML accuracy, while relative reduction in Test set error reaches more than 50% for higher baseline accuracy ML models. The two basic structures in the AL-MB, a competitive inhibition layer and a high-dimensional sparse layer, coupled with Hebbian plasticity, act as effective, automated feature generators that substantially improve ML classification in the test case we examine.