ESPRESSO: Entropy and Shape Aware Time Series Segmentation for Processing Heterogeneous Sensor Data
summary
ESPRESSO segments multivariate sensor streams by combining Matrix Profile pattern matching and an entropy-driven greedy search (inspired by IGTS) to pick boundaries that minimize segment entropy. It augments FLOSS with temporal chaining and weighted arc density, runs with a single subsequence-length parameter, and outperforms four baselines on seven datasets measured by RMSE and F-score. Ablations show shape and entropy components both contribute, and two case studies cover daily routine deviation and emotion inference.
AI-written summary · about
related talks
35mSignatures, patterns and trends: Timeseries data mining @ Etsy
55mSearching for Time Series Patterns at Big Data Scale
44mMethods and Tools for Time Series Data Science Problems
30mExtracting insights from unstructured data
01h 29msktime: time series anomaly detection, changepoint detection, segmentation
31mAdvanced machine learning time series....
17mFrom Sensor to Value: A journey
31mML time series forecasting methods