FreqRISE: Explaining time series using frequency masking: Thea Brüsch (DTU Compute) 

Thea Brüsch, a PhD student at the Technical University of Denmark (DTU Compute) and guest researchers visiting Visual Intelligence at UiT, gave a presentation titled "FreqRISE: Explaining time series using frequency masking" on June 20th 2024 as part of the Visual Intelligence Online Seminar series.
Abstract:
Time series data is fundamentally important for describing many critical domains such as healthcare, finance, and climate, where explainable models are necessary for safe automated decision-making. To develop eXplainable AI (XAI) in these domains therefore implies explaining salient information in the time series. Current methods for obtaining saliency maps directly build on methods developed for computer vision and assume localized information in the raw input space. However, for many time series the salient information is more likely to be localized in the frequency or time-frequency domain. In this talk, I will present FreqRISE, which transforms the time series to either spectrums or spectrograms and provides the explanations in these domains.

Наука