B-cos GNNs: Faithful Explanations through Dynamic Linearity
For graph neural network users requiring interpretable predictions, B-cos GNNs provide inherently faithful explanations without auxiliary explainers, though with a slight accuracy trade-off.
B-cos GNNs achieve faithful, instance-level explanations by decomposing predictions into per-node, per-feature contributions via dynamic linearity, trading small accuracy losses for state-of-the-art explainability and orders-of-magnitude faster explanation generation.
We introduce B-cos GNNs, an inherently explainable class of graph neural networks whose predictions decompose exactly into per-node, per-feature contributions via a single input-dependent linear map. B-cos GNNs use linear (sum-based) aggregation and replace non-linear message and update functions with B-cos transforms. This induces meaningful, task-specific weight-input alignment that is directly accessible through the model's dynamic linearity. Instance-level explanations follow from a single forward and backward pass, requiring no auxiliary explainer, modified learning objective, or perturbation procedure. Instantiated as a GIN, our approach trades small losses in predictive accuracy for state-of-the-art explainability across diverse synthetic and real-world benchmarks, producing explanations orders of magnitude faster than post-hoc baselines.