Improving Robustness of Spectrogram Classifiers with Neural Stochastic Differential Equations

ABSTRACT Signal analysis and classification is fraught with high levels of noise and perturbation. Computer-vision-based deep learning models applied to spectrograms have proven useful in the field of signal classification and detection; however, these methods aren't designed to handle the low signalto-noise Figure 1: Surface representations of the 2D Brownian surface ratios inherent within non-vision signal processing noise injected into our Neural SDE tasks. While they are powerful, they are currently not the method of choice in the inherently noisy and dynamic critical confidence of neural networks in their responses. Such confidence infrastructure domain, such as smart-grid sensing, anomaly metrics will enable human Subject Matter Experts to detection, and non-intrusive load monitoring. Currently, build a relationship of trust with robust neural networks that these models can be brittle, which makes them susceptible to have a history of credible and correctly calibrated responses.