Reproduce “What-If Explanations Over Time: Counterfactuals for Time Series Classification” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Udo Schlegel, Thomas Seidl

Udo Schlegel and Thomas Seidl from LMU Munich present a structured review of counterfactual explanation (CFE) algorithms for time series classification, categorizing existing methods into six families. They introduce the open-source Counterfactual Explanations for Time Series (CFTS) library to standardize implementation and evaluation, highlighting how different approaches address challenges like temporal coherence and plausibility in time series data.

Key insights

  • Time series counterfactuals require specialized approaches to maintain temporal coherence and plausibility, moving beyond simple distance minimization in input space.
  • Different CFE methodological families exhibit distinct trade-offs between criteria such as validity, proximity, sparsity, and temporal plausibility.
  • Standardized evaluation using temporal-specific metrics (e.g., DTW, spectral similarity) is crucial for accurately assessing the quality and realism of time series counterfactuals.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.