REALITrees: Rashomon Ensemble Active Learning for Interpretable Trees

TL;DR

REALITrees introduces a novel active learning method that constructs a committee of near-optimal models from the Rashomon set, improving sample efficiency and convergence in noisy environments for interpretable decision trees.

Contribution

It proposes Rashomon Ensembled Active Learning (REAL) that exhaustively enumerates the Rashomon set for sparse decision trees, enhancing active learning performance.

Findings

Outperforms randomized ensembles in active learning tasks.

Achieves faster convergence in noisy environments.

Leverages model multiplicity for improved sample efficiency.

Abstract

Active learning reduces labeling costs by selecting samples that maximize information gain. A dominant framework, Query-by-Committee (QBC), typically relies on perturbation-based diversity by inducing model disagreement through random feature subsetting or data blinding. While this approximates one notion of epistemic uncertainty, it sacrifices direct characterization of the plausible hypothesis space. We propose the complementary approach: Rashomon Ensembled Active Learning (REAL) which constructs a committee by exhaustively enumerating the Rashomon Set of all near-optimal models. To address functional redundancy within this set, we adopt a PAC-Bayesian framework using a Gibbs posterior to weight committee members by their empirical risk. Leveraging recent algorithmic advances, we exactly enumerate this set for the class of sparse decision trees. Across synthetic and established active learning baselines, REAL outperforms randomized ensembles, particularly in moderately noisy environments where it strategically leverages expanded model multiplicity to achieve faster convergence.