Reduan Achtibat and Maximilian Dreyer, two PhD students at Fraunhofer Heinrich Hertz Institute Berlin, gave a talk titled "Towards Explainable AI 2.0 with Concept-based Explanations" on May 28th 2024.
The talk is divided into two parts:
0:00: Concept Relevance Propagation (CRP)
16:38: Prototypical Concept-based Explanations (PCX)
Abstract:
While local XAI methods explain individual predictions in form of attribution maps, thereby identifying where important features occur (but not providing information about what they represent), global explanation techniques visualize what concepts a model has generally learned to encode. Both types of methods thus only provide partial insights and leave the burden of interpreting the model's reasoning to the user. In this work we introduce the Concept Relevance Propagation (CRP) approach, which combines the local and global perspectives and thus allows answering both the "where" and "what" questions for individual predictions.
Concept-based explanations allow for a detailed understanding of deep neural networks. However, studying a model based on single explanations can be unfeasible for large datasets. With our recent Prototypical Concept-based Explanations (PCX) method, we propose to summarize similar explanation via prototypes. As such, we can understand the whole model behavior quickly and in detail. PCX further allows to validate individual predictions by communicating the differences to the ordinary behavior (as given by the prototypes). By quantifying these differences, we can assign new predictions to known prototypes, or detect outlier predictions.