Signal Temporal logic Embedding for Logically-grounded Learning and Explanation

STELLE is a neuro-symbolic framework for interpretable time series classification.
It unifies classification and explanation through direct embedding of trajectories into a space of Signal Temporal Logic (STL) concepts.

Each prediction made by STELLE is accompanied by human-readable logical explanations, both:

• Local — explaining individual predictions
• Global — describing class-level temporal behaviour

The method was introduced in:

Irene Ferfoglia, Simone Silvetti, Gaia Saveri, Laura Nenzi, and Luca Bortolussi.
Guided by Stars: Interpretable Concept Learning Over Time Series via Temporal Logic Semantics
submitted to Journal of Artificial Intelligence Research (JAIR), 2025.

• Interpretable-by-design classification using STL formulae
• Trajectory embedding kernel based on temporal robustness
• Dual explanations — local and global
• Compatible with multivariate time series
• Implemented in PyTorch

If you use STELLE in your research, please cite:

@article{ferfoglia2025stelle,
  title     = {Guided by Stars: Interpretable Concept Learning Over Time Series via Temporal Logic Semantics},
  author    = {Ferfoglia, Irene and Silvetti, Simone and Saveri, Gaia and Nenzi, Laura and Bortolussi, Luca},
  year      = {2025}
}