Label-Descriptive Patterns and Their Application to Characterizing Classification Errors
Preprint
Abstract
State-of-the-art deep learning methods achieve human-like performance on many tasks, but make errors nevertheless. Characterizing these errors in easily interpretable terms gives insight into whether a classifier is prone to making systematic errors, but also gives a way to act and improve the classifier. We propose to discover those feature-value combinations (i.e., patterns) that strongly correlate with correct resp. erroneous predictions to obtain a global and interpretable description for arbitrary classifiers. We show this is an instance of the more general label description problem, which we formulate in terms of the Minimum Description Length principle. To discover a good pattern set, we develop the efficient Premise algorithm. Through an extensive set of experiments we show it performs very well in practice on both synthetic and real-world data. Unlike existing solutions, it ably recovers ground truth patterns, even on highly imbalanced data over many features. Through two case studies on Visual Question Answering and Named Entity Recognition, we confirm that Premise gives clear and actionable insight into the systematic errors made by modern NLP classifiers.Keywords
All Related Versions
This publication has 12 references indexed in Scilit:
- Interpretable multiclass classification by MDL-based rule listsInformation Sciences, 2020
- Evaluating Models’ Local Decision Boundaries via Contrast Sets2020
- Automated Data Slicing for Model Validation: A Big Data - AI Integration ApproachIEEE Transactions on Knowledge and Data Engineering, 2019
- An overview of emerging pattern mining in supervised descriptive rule discovery: taxonomy, empirical study, trends, and prospectsWIREs Data Mining and Knowledge Discovery, 2017
- A peek into the black box: exploring classifiers by randomizationData Mining and Knowledge Discovery, 2014
- Diverse subgroup set discoveryData Mining and Knowledge Discovery, 2012
- Kingfisher: an efficient algorithm for searching for both positive and negative dependency rules with statistical significance measuresKnowledge and Information Systems, 2011
- A linear-time algorithm for computing the multinomial stochastic complexityInformation Processing Letters, 2007
- A Universal Prior for Integers and Estimation by Minimum Description LengthThe Annals of Statistics, 1983
- Modeling by shortest data descriptionAutomatica, 1978