8月演講

Analysis Seminar, 吳浩榳教授, Department of Mathematics Courant Institute of Mathematical Sciences, New York University Monday, August 5, 11:10—12:30 數學系3樓會議室 Title: Nonstationary Time Series Analysis with Manifold Learning Techniques Abstract: Spectral embedding is a popular dimension reduction technique, yet scaling it to handle large datasets and ensuring robustness remain challenges. In response, we introduce Robust and Scalable Embedding via Landmark Diffusion (ROSELAND). This method leverages a small set of landmarks to efficiently compute spectral embeddings, generalizing the diffusion map (DM) algorithm while preserving its key properties. We demonstrate ROSELAND's scalability and robustness, provide theoretical insights into its asymptotic behavior, and validate its performance through simulations.

Analysis Seminar, 吳浩榳教授, Department of Mathematics Courant Institute of Mathematical Sciences, New York University Friday, August 9, 15:00—16:00 數學系3樓會議室 Title: A Tutorial on Single Channel Blind Source Separation We will discuss the single channel blind source separation problem. The focus will be the application of optimal shrinkage and random matrix theory to achieve the decomposition. The problem of extracting fetal electrocardiogram from the single channel trans-abdominal electrocardiogram will be demonstrated as an application. Its relationship with manifold denoising will also be discussed.