PSEUDo: Interactive Pattern Search in Multivariate Time Series with Locality-Sensitive Hashing and Relevance Feedback

Abstract

We present PSEUDo, a visual pattern retrieval tool for multivariate time series. It aims to overcome the uneconomic (re- )training with deep learning-based methods. Very high-dimensional time series emerge on an unprecedented scale due to increasing sensor usage and data storage. Visual pattern search is one of the most frequent tasks on such data. Automatic pattern retrieval methods often suffer from inefficient training, a lack of ground truth, and a discrepancy between the similarity perceived by the algorithm and the user. Our proposal is based on a query-aware locality-sensitive hashing technique to create a representation of multivariate time series windows. It features sub-linear training and inference time with respect to data dimensions. This performance gain allows an instantaneous relevance-feedback-driven adaption and converges to users' similarity notion. We are benchmarking PSEUDo in accuracy and speed with representative and state-of-the-art methods, evaluating its steerability through simulated user behavior, and designing expert studies to test PSEUDo's usability.