- Date: 16 Dec. (Thu.)
- Place: West Wing, 6th floor, Colloquium Room and on the Web (zoom)
- Time: 16:50–18:20
- Speaker : Hau-Tieng Wu (New York University, Courant)
- Title: Manifold denoising for Nonstationary Biomedical Time Series with Neuromodulation Applications
- Abstract:
Recent advances in technologies enable continuous acquisition of high frequency and multimodal physiological waveforms, moving well beyond traditional pointwise or sparse clinical measurements. These data streams are usually represented as nonstationary time series, often exhibiting complicated time-varying periodic structure and nonlinear dynamics, which pose fundamental challenges for statistical modeling and machine learning. To handle such time series, we introduce a manifold-denoising based data sharpening technique that processes raw nonstationary time series by converting it into manifold-valued point cloud for learning. The proposed technique leverages random matrix theory and spectral geometry to achieve manifold denoising, and hence time series analysis, and has rigorous theoretical guarantees. In addition to open problems, a clinical application in real-time neuromodulation will be presented to illustrate how this framework improves artifact suppression and signal interpretation, highlighting its potential to enhance physiological data analysis and support medical decision-making.
日本語
English