Computer Science Colloquium - Rodica Neamtu

Interactive Exploration of Time Series Powered by Time Warped Distances

Modern applications in this digital age collect a staggering amount of data from diverse domains ranging from astronomy, finance, and e-commerce to genome sequencing, a significant percentage of which is in the form of time series. To make sense of it, analysts interactively sift through these collections in search of critical relationships between and recurring patterns within these time series. In this talk I will show how we can answer complex questions by using effective exploratory analysis of time series datasets composed of heterogeneous, variable-length and misaligned time series.

I will first describe how our novel paradigm ONEX extracts relevant time series matches based on the compute-intensive yet robust alignment Dynamic Time Warping distance from a simple-to-compute Euclidean-prepared base. This strategy leads to highly accurate results that are faster than the fastest-known so far state-of-the-art DTW-based system. We extend our formal model interleaving the inexpensive Euclidean distance with the DTW to find similar sequences of any length and temporal alignment with response times that are almost as fast as retrieving only one best match.

Second, I will discuss our conceptual framework that supports alignment (warping) of a large array of domain-specific distances. This is the first work to generalize the ubiquitous DTW distance and ÒextendÓ its warping capabilities to a diverse set of existing point-to-point distances covering a wide range of popular functions. Our extensive experiments on real datasets reveal that some of the newly warped distances are even better than the classic DTW for data mining tasks such as classification and clustering.

Rodica Neamtu is an Associate Teaching Professor of Computer Science and Data Science at Worcester Polytechnic Institute. With more than 15 years of teaching experience in various academic institutions, she works with graduate and undergraduate students on research to add expressive exploratory mechanisms to big time series collections. Her main research interests are at the confluence of theoretical computer science, data mining, and Big Data. Her work contributes to developing and leveraging groundbreaking techniques for mining time series datasets. She focuses on exploring the theoretical underpinnings of these methods, as well as the practical issues at the heart of Big Data.