Research

Research on evolving data combines time-series modeling, tensor methods, and graph learning to understand patterns, relationships, and future behavior.

Tensor Forecasting

Tensor forecasting A small tensor cube represents several data dimensions. Beside it, a solid time series continues into a dotted future prediction. Past Future
How will multi-aspect data evolve?

Tensor models forecast how activity evolves across time, locations, and other attributes. By capturing shared dynamics, trends, and seasonality, these models connect future predictions with interpretable patterns in multidimensional data.

Dependency Discovery

Dependency discovery Three observed signals are paired with a sparse network. Undirected lines show selected statistical relationships, without implying causality. Signals Links
Which variables are related?

Dependency discovery reveals how statistical relationships change across multivariate and tensor time series. Structured network models separate relationships across data dimensions, making it possible to inspect which variables are connected and how those connections evolve.

Anomaly Detection

Anomaly detection A group of similar observations forms a soft cluster. One unusual observation is highlighted in coral with a surrounding ring. Expected Unusual
Which observed behaviors are unusual?

Anomaly detection methods identify unusual behavior in event streams and interaction graphs. Temporal patterns, node roles, and heterogeneous attributes provide context for finding deviations and interpreting their characteristics, with applications in cybersecurity and communications.

Predictive Maintenance (PdM)

Predictive maintenance A sensor and its observations lead into a dotted future trajectory. The interval between now and a future event represents the time to be predicted. Now Event
How much time remains until a failure?

Predictive maintenance uses sensor data to forecast equipment events and estimate the time remaining until failure. Models capture changing operating stages and relationships between sensors, updating predictions as new measurements arrive.

View all publications