Multi-Objective Counterfactual Explanations

Dandl, Susanne; Molnar, Christoph; Binder, Martin; Bischl, Bernd

doi:10.1007/978-3-030-58112-1_31

Cited by 170 publications

(138 citation statements)

References 24 publications

Supporting

Mentioning

136

Contrasting

Unclassified

Order By: Relevance

“…Others (e.g., onehot encoding) mitigate the computational problem at the expense of neglecting the complex relationships between and within categorical variables [10]. Recently, researchers have proposed to utilize genetic algorithms to generate counterfactual explanations [23,24]. While capable of generating foils for mixed data, these approaches neither capture the full complexity of categorical variables nor effectively address desired characteristics of explanations.…”

Section: Methods For the Generation Of Coherent Counterfactual Explanmentioning

confidence: 99%

Capturing Users’ Reality: A Novel Approach to Generate Coherent Counterfactual Explanations

Förster

Hühn²,

Klier

et al. 2021

Proceedings of the Annual Hawaii International Conference on System Sciences

View full text Add to dashboard Cite

The opacity of Artificial Intelligence (AI) systems is a major impediment to their deployment. Explainable AI (XAI) methods that automatically generate counterfactual explanations for AI decisions can increase users' trust in AI systems. Coherence is an essential property of explanations but is not yet addressed sufficiently by existing XAI methods. We design a novel optimizationbased approach to generate coherent counterfactual explanations, which is applicable to numerical, categorical, and mixed data. We demonstrate the approach in a realistic setting and assess its efficacy in a humangrounded evaluation. Results suggest that our approach produces explanations that are perceived as coherent as well as suitable to explain the factual situation.

show abstract

Section: Methods For the Generation Of Coherent Counterfactual Explanmentioning

confidence: 99%

Capturing Users’ Reality: A Novel Approach to Generate Coherent Counterfactual Explanations

Förster

Hühn²,

Klier

et al. 2021

Proceedings of the Annual Hawaii International Conference on System Sciences

View full text Add to dashboard Cite

show abstract

“…As we have seen, the seminal work on CF explanation [Wacheter et al, 2018;Mittelstadt et al, 2019] proposes perturbing the features of synthetic CF instances, under a loss function balancing proximity to the test-instance against proximity to the decision boundary for the CF class, using a scaled L1-norm distance-metric. This idea has inspired follow-on work using different distance metrics (e.g., L2-norm) or, indeed, combinations of distance metrics [Dandl et al, 2020;Artelt and Hammer, 2020], with added constraints to deliver diverse CFs [Mothilal et al, 2020;. Hence, later, we will argue for the use of selected distance metrics to benchmark evaluations (ideally, ones that are psychologically grounded).…”

Section: Counterfactual Insightsmentioning

confidence: 99%

“…Laugel et al [2019] showed that one type of "bad" CF (i.e., out-of-distribution items) can be as high as 36% for some CF methods and Delaney et al [2020] have shown that even close, low-sparcity CFs can be out-of-distribution (see Figure 2). In the 100 systems reviewed here, we found that only 22% report "coverage results", though the definitions of the concept differ [Keane and Smyth, 2020;Schleich et al, 2021;Dandl et al, 2020].…”

Section: Deficit #4: Covering Coveragementioning

confidence: 99%

If Only We Had Better Counterfactual Explanations: Five Key Deficits to Rectify in the Evaluation of Counterfactual XAI Techniques

Keane

Kenny

Delaney

et al. 2021

Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence

View full text Add to dashboard Cite

In recent years, there has been an explosion of AI research on counterfactual explanations as a solution to the problem of eXplainable AI (XAI). These explanations seem to offer technical, psychological and legal benefits over other explanation techniques. We survey 100 distinct counterfactual explanation methods reported in the literature. This survey addresses the extent to which these methods have been adequately evaluated, both psychologically and computationally, and quantifies the shortfalls occurring. For instance, only 21% of these methods have been user tested. Five key deficits in the evaluation of these methods are detailed and a roadmap, with standardised benchmark evaluations, is proposed to resolve the issues arising; issues, that currently effectively block scientific progress in this field.

show abstract

“…data distribution and the resulting counterfactual are connected via high-density paths to the explained instance. Dandl et al [82] proposed a Multi-Objective Counterfactuals (MOC) method which translates the counterfactual search into a multi-objective optimization problem. The proposed approach returns a diverse set of counterfactuals with different trade-offs between the proposed objectives and maintains diversity in the feature space.…”

Section: ) Model-agnostic Counterfactual Explanationsmentioning

confidence: 99%

Explainable Artificial Intelligence for Tabular Data: A Survey

2021

View full text Add to dashboard Cite

Machine learning techniques are increasingly gaining attention due to their widespread use in various disciplines across academia and industry. Despite their tremendous success, many such techniques suffer from the "black-box" problem, which refers to situations where the data analyst is unable to explain why such techniques arrive at certain decisions. This problem has fuelled interest in Explainable Artificial Intelligence (XAI), which refers to techniques that can easily be interpreted by humans. Unfortunately, many of these techniques are not suitable for tabular data, which is surprising given the importance and widespread use of tabular data in critical applications such as finance, healthcare, and criminal justice. Also surprising is the fact that, despite the vast literature on XAI, there are still no survey articles to date that focus on tabular data. Consequently, despite the existing survey articles that cover a wide range of XAI techniques, it remains challenging for researchers working on tabular data to go through all of these surveys and extract the techniques that are suitable for their analysis. Our article fills this gap by providing a comprehensive and up-to-date survey of the XAI techniques that are relevant to tabular data. Furthermore, we categorize the references covered in our survey, indicating the type of the model being explained, the approach being used to provide the explanation, and the XAI problem being addressed. Our article is the first to provide researchers with a map that helps them navigate the XAI literature in the context of tabular data. INDEX TERMSBlack-box models, Explainable Artificial Intelligence, Machine Learning, Model interpretability VOLUME 4, 2016This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3116481, IEEE Access M. Sahakyan et al.: Explainable Artificial Intelligence for Tabular Data: A Survey• Decomposability, reflecting the degree to which each input, parameter, and calculation can be explained intuitively;• Algorithmic transparency, reflecting the degree to which the inner workings of the learning algorithm can be understood.For example, rule-based models [16] are considered transparent since they use a series of if-then rules that can easily be understood without the need for any further explanation. Unlike transparent models, black-box models are those that do not explain their predictions in a way that humans can understand [17]. Examples of black-box models include artificial neural networks [18] and gradient boosting [19]. Although black-box models are hard to interpret by humans, they tend to have higher prediction accuracy compared to their transparent counterparts. This trade-off between accuracy and transparency gives raise to the black-box explanation problem, which involves explaining the rationale behind the decisions made by black-box models. By providing such explanations, one can continue ...

show abstract

Multi-Objective Counterfactual Explanations

Cited by 170 publications

References 24 publications

Capturing Users’ Reality: A Novel Approach to Generate Coherent Counterfactual Explanations

Capturing Users’ Reality: A Novel Approach to Generate Coherent Counterfactual Explanations

If Only We Had Better Counterfactual Explanations: Five Key Deficits to Rectify in the Evaluation of Counterfactual XAI Techniques

Explainable Artificial Intelligence for Tabular Data: A Survey

Contact Info

Product

Resources

About