Learnable Masks for Time Series Explainability using Time-Frequency Representations

Theresa Dahl Frehr, Thea Brüsch, Tommy Sonne Alstrøm

Keywords: Time Series, Explainability, Time-Frequency Representations

TL;DR: Time–frequency saliency maps via DWT and STFT provide faithful time series explanations.

Abstract: The demand for explainable AI models continues to grow with the rise of AI-based solutions. Relatively few explainability methods address time series due to the complex nature of the data. We propose learning saliency maps over both time and frequency via the discrete wavelet transform and the short-time Fourier transform. Faithfulness scores show that our method is on par with current state-of-the-art methods.

Serve As Reviewer: ~Tommy_Sonne_Alstrøm1

Submission Number: 37