Visual Explanations via Iterated Integrated Attributions
This addresses the need for better interpretability in vision AI, providing a novel method for generating accurate explanations, though it appears incremental as it builds on existing attribution techniques.
The paper tackles the problem of explaining predictions in vision models by introducing Iterated Integrated Attributions (IIA), a generic method that uses iterative integration across inputs, representations, and gradients to produce precise explanation maps, and it demonstrates that IIA outperforms other state-of-the-art techniques in evaluations across tasks, datasets, and architectures.
We introduce Iterated Integrated Attributions (IIA) - a generic method for explaining the predictions of vision models. IIA employs iterative integration across the input image, the internal representations generated by the model, and their gradients, yielding precise and focused explanation maps. We demonstrate the effectiveness of IIA through comprehensive evaluations across various tasks, datasets, and network architectures. Our results showcase that IIA produces accurate explanation maps, outperforming other state-of-the-art explanation techniques.