Estimation of stimuli timing to evaluate chemical plume tracing behavior of the silk moth

Abstract

Insects serve as ideal models for replicating the adaptability of biological systems in Chemical Plume Tracing (CPT) because they perform efficient olfactory tracking. In this paper, we propose to evaluate the CPT behavior of the silk moth (Bombyx Mori) from the perspective of machine learning. We use a classification approach consisting of the Gaussian Mixture Model with Expectation Maximization (GMMEM) and the Echo State Networks (ESN) to identify the initial motion phases upon stimulation. The former method classifies the locomotion observation consisting of the linear and angular velocity into Gaussian density components which represent different elemental motions. Then, these motions are used as training data for the ESN to estimate the initial motion phases upon stimulation which represents the stimuli timing. The same procedure is implemented on different moths and cross-evaluation is done among the moths in the sample to evaluate their behavior singularity. This method achieves decent estimation accuracy and serves as a feasible approach to complement the conventional neurophysiology analysis of insects behavior. The results also suggest the presence of CPT behavior singularity for silk moths.