Weakly Supervised Multi-task Learning for Concept-based Explainability

In ML-aided decision-making tasks, such as fraud detection or medical diagnosis, the human-in-the-loop, usually a domain-expert without technical ML knowledge, prefers high-level concept-based explanations instead of low-level explanations based on model features. To obtain faithful concept-based explanations, we leverage multi-task learning to train a neural network that jointly learns to predict a decision task based on the predictions of a precedent explainability task (i.e., multi-label concepts). There are two main challenges to overcome: concept label scarcity and the joint learning. To address both, we propose to: i) use expert rules to generate a large dataset of noisy concept labels, and ii) apply two distinct multi-task learning strategies combining noisy and golden labels. We compare these strategies with a fully supervised approach in a real-world fraud detection application with few golden labels available for the explainability task. With improvements of 9.26% and of 417.8% at the explainability and decision tasks, respectively, our results show it is possible to improve performance at both tasks by combining labels of heterogeneous quality. Figure 1: Weakly supervised multi-task learning strategies for concept-based explainability: (A) baseline strategy resorts exclusively to golden explainability labels; (B) two-stage learning strategy (1) uses noisy explainability labels to pre-train a base model and (2) fine-tuning either using purely golden batches or hybrid ones; (C) hybrid learning strategy only uses hybrid batches of golden and noisy explainability labels. The AI black-box paradigm has led to a growing demand for model explanations (Ribeiro et al., 2016; Lundberg & Lee, 2017). It concerns the generation of high-level concept-based explanations (e.g., "Suspicious payment") rather than low-level explanations based on model features (e.g., "MCC 7801"). Concept-based explainability can be implemented using a multi-task learning approach (Kim et al., 2018; Melis & Jaakkola, 2018; Ghorbani et al., 2019; Koh et al., 2020).