Incremental SliceLine for Iterative ML Model Debugging under Updates

Abstract

SliceLine is a model debugging technique for finding the top-K worst slices (in terms of conjunctions of attributes) where a trained machine learning (ML) model performs significantly worse than on average. In contrast to other slice finding techniques, our prior work SliceLine introduced an intuitive scoring function, effective pruning strategies, and fast linear-algebra-based evaluation strategies. Together, SliceLine is able to find the exact top-K worst slices in the full lattice of possible conjunctions of attributes in reasonable time. Recently, we observe a major trend towards iterative algorithms that incrementally update the dataset (eg, selecting samples, augmenting with new instances) and ML model. Fully computing SliceLine from scratch for every update is, however, unnecessarily wasteful. In this paper, we introduce an incremental problem formulation of SliceLine, new pruning strategies that leverage state of previous slice finding runs on a modified dataset, and an extended linear-algebra-based enumeration algorithm. Our experiments show that incremental SliceLine yields robust runtime improvements of up to an order of magnitude faster than full SliceLine, while still allowing effective parallelization in local, distributed, and federated environments.

Citation

Zoepffel, Frederic Caspar, Christina Dionysio, and Matthias Boehm. “Incremental SliceLine for Iterative ML Model Debugging under Updates.” In Datenbanksysteme für Business, Technologie und Web (BTW 2025), pp. 567-587. Gesellschaft für Informatik, Bonn, 2025.

@inproceedings{zoepffel2025incremental,
  title={Incremental SliceLine for Iterative ML Model Debugging under Updates},
  author={Zoepffel, Frederic Caspar and Dionysio, Christina and Boehm, Matthias},
  booktitle={Datenbanksysteme f{\"u}r Business, Technologie und Web (BTW 2025)},
  pages={567--587},
  year={2025},
  organization={Gesellschaft f{\"u}r Informatik, Bonn}
}

Links

• Paper