Examples are not enough, learn to criticize! Criticism for Interpretability

By Been Kim, Oluwasanmi Koyejo, Rajiv Khanna. 2016

867 citations recorded by the source corpus.

Subfields: Safety, Robustness & Interpretability

How researchers see this paper

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

In Safety, Robustness & Interpretability

Citations in the corpus

Reframing Human-AI Collaboration for Generating Free-Text Explanations (2021, background): “(not enough, enough, or too much) This question is designed to test the extent to which the provided novel informa-15 tion is adequate or sufﬁcient (Kim et al., 2016; Lei et al., 2016; Ehsan et al., 2019).”
Measuring Association Between Labels and Free-Text Rationales (2020, background): “…2017) has a rich history in machine learning inter-pretability as a reliable measure of rationale quality from the lens of utility to an end-user (Kim et al., 2016; Chandrasekaran et al., 2018; Hase and Bansal, 2020; Yeung et al., 2020; Poursabzi-Sangdeh et al., 2021; Rajagopal et al., 2021,…”
Interpretable Machine Learning: Fundamental Principles and 10 Grand Challenges (2021, methodology): “Kim et al. (2016) formulated the prototype learning problem as an optimization program that minimizes the squared maximum mean discrepancy, which is a submodular optimization problem and can be solved approximately using a greedy algorithm.”
Kernel Mean Embedding of Distributions: A Review and Beyonds (2016, background): “criticism (Lloyd and Ghahramani 2015, Kim et al. 2016), approximate Bayesian computation (Park et al. 2016), and model selection in generative models (Bounliphone et al. 2016), for example.”
FairSight: Visual Analytics for Fairness in Decision Making (2019, background): “Instance-level explanations that identify instances such as counterfactual examples or prototypes [24] seek to generate an explanation with a single instance.”
Generative Counterfactual Introspection for Explainable Deep Learning (2019, methodology): “A promising technique for investigating decision boundaries of a model is based on the prototype and criticism based explanations approach [9].”
Explanation in Human-AI Systems: A Literature Meta-Review, Synopsis of Key Ideas and Publications, and Bibliography for Explainable AI (2019, background): “…Halpern and Pearl (2005a,b), Hilton, McClure and Slugoski (2005); Hitchcock (2003); Hitchcock and Woodward (2003); Hoffman, et al., (2014); Khemlani and Johnson-Laird (2010); Kim, Khanna, and Koyejo (2016); Lipton, (1990); Mandel (2003a,b), McCloy and Byrne (2000); and Thompson and Byrne (2002).”
Peeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI) (2018, methodology): “Kim [136] developed an unsupervised algorithm for automatically finding prototypes and critics for a dataset, called MMD-critic.”
A Survey of Methods for Explaining Black Box Models (2018, background): “us types of probability distributions which are assumed to be correct for each type of data. [42] proposes an extension of BCM which exploits humans interaction to improve the prototypes. Finally, in [43] the approach is further expanded to include criticisms, where a criticism is an instance that d”
Explainable machine learning in deployment (2019, background): “se techniques attempt to characterize the concepts learned by the model [32], simpler models learned from the representation of complex models [19], prototypical samples from a particular model output[14], or the topology of the data itself [21]. None of our interviewees reported deploying global ex”
Deep Neural Decision Trees (2018, background): “, 2017), nearest neighbour models (Kim et al., 2016), and neural networks (Kim et al.”
Interpretable 3D Human Action Analysis with Temporal Convolutional Networks (2017, background): “The work of [18] pushes this idea further by forcing a model to produce both exemplars and criticisms.”
ConvNets and ImageNet Beyond Accuracy: Explanations, Bias Detection, Adversarial Examples and Model Criticism (2017, methodology): “Recently, Kim et al. [24] have proposed to use, in addition to the prototypes, data points sampled from regions of the input space not well captured by the model or the prototypes.”
Explaining machine learning classifiers through diverse counterfactual explanations (2019, background): “k focuses on. 2.3 Explanation through Examples The most relevant class of explanations to our approach is through examples. An example-based explanation framework is MMD-critic proposed by Kim et al. [18], which selects both prototypes and criticisms from the original data points. More recently, cou”
ProtoDash: Fast Interpretable Prototype Selection (2017, methodology): “Interpretable modeling has received a lot of attention in recent times [17, 12, 10, 20, 19].”
From Anecdotal Evidence to Quantitative Evaluation Methods: A Systematic Review on Evaluating Explainable AI (2022, background): “Prototypes (Parts of) Representative examples, including concepts [127], influential training instances [90], prototypical parts [36, 179], nearest neighbors and criticisms [125].”
Explainable, trustworthy, and ethical machine learning for healthcare: A survey (2021, background): “[36] as the extent to which humans can continuously predict results from the model.”