Ridge analysis of time-frequency representations in the presence of Noise

Data Science Seminar

Hau-Tieng Wu
NYU

Abstract

The advancement of wearable devices has provided physicians with extensive diagnostic insights for healthcare. While time-frequency analysis is widely used to examine nonstationary time series in clinical applications, it remains challenging, particularly in the presence of noise. In this talk, I will discuss recent progress in addressing a key yet often overlooked step, ridge analysis, under a noise model, extending the Borell-TIS inequality and Dudley’s theorem to time-frequency representations. Ridge analysis serves as a bridge between time-frequency analysis, signal processing, and statistical inference, playing a crucial role in extracting reliable and actionable clinical information on a strong scientific foundation. The assessment of photoplethysmogram signal quality in digital health will be presented as an example.