Class-Conditional Conformal Prediction for Signal Temporal Logic Inference with Many Classes
The work addresses a gap where accurate time series classification does not guarantee that the derived Signal Temporal Logic (STL) formulas correctly reflect the true class structure in multiclass settings. The method introduces a two-stage procedure: first, learning discriminative formulas, and second, refining threshold parameters to ensure formula satisfaction while maintaining class separation. This results in finite-sample class-conditional coverage guarantees.