Towards Automatic Concept-based Explanations

By Amirata Ghorbani, James Wexler, James Y. Zou, Been Kim. 2019

833 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI (2019, methodology): “ACE algorithm [55] uses TCAV to compute saliency score and generate super-pixels as explanations.”
Concept Bottleneck Models (2020, methodology): “Many methods have been developed to interpret models post-hoc, including recent work on using human-specified concepts to generate explanations (Bau et al., 2017; Kim et al., 2018; Zhou et al., 2018; Ghorbani et al., 2019).”
Concept whitening for interpretable image recognition (2020, methodology): “hods try to learn a linear combination of them to represent a predeﬁned concept (Kim et al.,2017) or to automatically discover concepts by clustering patches and deﬁning the clusters as new concepts (Ghorbani et al., 2019). Although these methods are promising, they are based on assumptions of the l”
Formalizing Trust in Artificial Intelligence: Prerequisites, Causes and Goals of Human Trust in AI (2020, background): “fluence functions [38], probing [16] Adaptable user-centered explainability Factsheets (explainability) Counterfactual [21], contrastive [51], natural language [27], by-example [38], and conceptlevel [19] explanations Make AI systems identifiable as non-human - - Diversity, non-discrimination, fairn”
Trustworthy Artificial Intelligence: A Review (2022, background): “In-the-loop 3 Interpretable Model Global [40, 87, 120] Feature Importance Local [161] Global, Local [9, 66, 77, 93, 127, 227]”
A comprehensive taxonomy for explainable artificial intelligence: a systematic survey of surveys on methods and concepts (2021, methodology): “The concept mining approach ACE [51] combines prototypes with examples: They search a concept embedding as prototype for an automatically collected set of example patches, that can be used to explain the prototype. feature importance that will highlight features with high attribution or influence on”
A Multidisciplinary Survey and Framework for Design and Evaluation of Explainable AI Systems (2018, background): “cept. They later extended their concept-based global explanation of model representation learning for systematic discovery of concepts that are human meaningful and important for the model prediction [63]. They used human subject experiments to evaluate learned concepts. Table 7 summarizes a list of”
Symbols as a Lingua Franca for Bridging Human-AI Chasm for Explainable and Advisable AI Systems (2021, unknown): “…note here that the previous works that try to learn symbols from the perspective of the system (cf. (Konidaris, Kaelbling, and Lozano-Perez 2018), (Bonet and Geffner 2019), (Ghorbani et al. 2019)) need not result in useful symbolic interfaces, as those learned symbols may not make sense to humans.”
Acquisition of chess knowledge in AlphaZero (2021, methodology): “Concept-based interpretability Concept-based interpretability methods build on the idea of network probing [19], and use human understandable concepts to explain neural network decisions in terms that users can understand [7, 8, 9, 10, 11].”
Post-hoc Interpretability for Neural NLP: A Survey (2021, background): “However, there is new research towards discovering concepts automatically [43].”
Opportunities and Challenges in Explainable Artificial Intelligence (XAI): A Survey (2020, methodology): “Newer methods proposed in literature such as explaining with Concepts [51, 52, 58] which we discussed in subsection”
Explainable artificial intelligence: an analytical review (2021, background): “Subsequently, automatic concept based explanations (Ghorbani et al., 2019) extract CAVs automatically without human supervision, thereby removing human bias.”
On the (In)fidelity and Sensitivity for Explanations. (2019, background): “Conceptbased explanation methods seek to explain the decision of the model by high-level human concepts [18, 14, 6].”
The alignment problem from a deep learning perspective (2022, background): “…et al., 2020, Wang et al., 2022, Elhage et al., 2021]; and conceptual interpretability, which aims to develop automatic techniques for probing and modifying human-interpretable concepts in networks [Ghorbani et al., 2019, Alvarez Melis and Jaakkola, 2018, Burns et al., 2022, Meng et al., 2022].”
Post-hoc Concept Bottleneck Models (2022, background): “The concepts in the library can be selected by a domain expert or learned automatically from the data (Ghorbani et al., 2019; Yeh et al., 2020).”
Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task (2022, background): “Finally, latent saliency maps can be viewed as a generalization of the TCAV approach Kim et al. (2018); Ghorbani et al. (2019); Koh et al. (2020).”
On Completeness-aware Concept-Based Explanations in Deep Neural Networks (2019, methodology): “For each part of data point x t (We omit i for notational simplicity), The inner product between the data and concept vector is viewed as the closeness of the input x t and the concept c following [5, 7].”