C. Molero-Río, B. Li, T. Wang, C. Rudin
Constructing a Sparse Rule Set (SRS) efficiently is one of the fundamental problems of interpretable machine learning. A rule set is a model of the form: If X satisfies (condition A AND condition B) OR (condition C) OR ··· , then Y = 1. SRS models have the advantages of being understandable to human experts and robust to outliers. This work presents FastSRS, an efficient algorithm for learning SRSs that trades off accuracy and sparsity. The algorithm’s key advantage is a set of theoretical bounds that efficiently reduces the size of the search space. The algorithm produces models on the accuracy vs. sparsity frontier more consistently and efficiently than previous approaches, and scales better to larger datasets.
Keywords: Interpretable Machine Learning, Rule and Patter Mining, Heuristic Search
Scheduled
Ramiro Melendreras Award IV
November 8, 2023 10:10 AM
CC4: Room 2