BEARS Make Neuro-Symbolic Models Aware of their Reasoning Shortcuts

TL;DR

This paper introduces BEARS, an ensembling method that makes neuro-symbolic models aware of reasoning shortcuts by calibrating their confidence, thereby improving their reliability and interpretability without sacrificing accuracy.

Contribution

The paper proposes BEARS, a novel ensembling approach that enhances neuro-symbolic models' awareness of reasoning shortcuts through confidence calibration, reducing overconfidence in flawed concepts.

Findings

BEARS improves models' awareness of reasoning shortcuts.

The method maintains prediction accuracy while calibrating concept confidence.

Enhanced RS-awareness facilitates targeted data annotation for mitigation.

Abstract

Neuro-Symbolic (NeSy) predictors that conform to symbolic knowledge - encoding, e.g., safety constraints - can be affected by Reasoning Shortcuts (RSs): They learn concepts consistent with the symbolic knowledge by exploiting unintended semantics. RSs compromise reliability and generalization and, as we show in this paper, they are linked to NeSy models being overconfident about the predicted concepts. Unfortunately, the only trustworthy mitigation strategy requires collecting costly dense supervision over the concepts. Rather than attempting to avoid RSs altogether, we propose to ensure NeSy models are aware of the semantic ambiguity of the concepts they learn, thus enabling their users to identify and distrust low-quality concepts. Starting from three simple desiderata, we derive bears (BE Aware of Reasoning Shortcuts), an ensembling technique that calibrates the model's concept-level confidence without compromising prediction accuracy, thus encouraging NeSy architectures to be uncertain about concepts affected by RSs. We show empirically that bears improves RS-awareness of several state-of-the-art NeSy models, and also facilitates acquiring informative dense annotations for mitigation purposes.