APPLIED MATH SEMINAR

Name:   Neta Rabin, Tel-Aviv University

Title:     Detection and prediction of trends in dynamically evolving systems

When/where:   Tuesday, September 30, 4:15 p.m., Room 200 AKW

Abstract:          

Detecting and predicting the behavior of a dynamic system that evolves with time is an important task. In essence, the goal is to change the representation of the captured datasets, originally in a form involving a large number of variables which dynamically change, into a low-dimensional description using only a small number of free parameters. The new embedded representation describes the data in a faithful manner, by preserving some quantities of interest like local mutual distances in the reduced space.

In this talk, the diffusion framework is applied to two datasets. The first is a dataset that was collected by a performance monitor. The dimensionality reduction of the data into a lower space creates an informative and reliable learning set that separates between normal and abnormal system behavior. The second dataset contains electricity prices. The dynamic behavior of the electricity price curve is embedded into a low dimensional manifold that captures the change in electricity consumption hour by hour. This low dimensional manifold is a natural space for prediction of day-ahead prices.