Toward a Unified Framework for Debugging Concept-based Models
This work addresses debugging challenges for users of concept-based models, but it is incremental as it builds on existing strategies for model alignment.
The paper tackles interactive debugging of concept-based models by introducing a schema for identifying and prioritizing bugs in concepts and aggregation functions, along with a novel loss function for debugging the aggregation step.
In this paper, we tackle interactive debugging of "gray-box" concept-based models (CBMs). These models learn task-relevant concepts appearing in the inputs and then compute a prediction by aggregating the concept activations. Our work stems from the observation that in CBMs both the concepts and the aggregation function can be affected by different kinds of bugs, and that fixing these bugs requires different kinds of corrective supervision. To this end, we introduce a simple schema for human supervisors to identify and prioritize bugs in both components, and discuss solution strategies and open problems. We also introduce a novel loss function for debugging the aggregation step that generalizes existing strategies for aligning black-box models to CBMs by making them robust to how the concepts change during training.