Intelligible models for classification and regression

Lou, Yin; Caruana, Rich; Gehrke, Johannes

doi:10.1145/2339530.2339556

Cited by 370 publications

(298 citation statements)

References 16 publications

Supporting

Mentioning

297

Contrasting

Order By: Relevance

“…In this context, a system is considered interpretable if "a user cannot only see, but also study and understand how inputs are mathematically mapped to outputs" (Doran et al 2017), with regression models (Schielzeth 2010) or generalized additive models (Lou et al 2012) serving as examples. However, as discussed for instance in Vellido et al (2012), Rudin (2014), interpretability in these cases refers almost exclusively to a mathematical property of the models, allowing for a certain degree of knowledge extraction from the model and subsequent interpretation by domain experts.…”

Section: Comprehensibility and Explanation In Machine Learningmentioning

confidence: 99%

Ultra-Strong Machine Learning: comprehensibility of programs learned with ILP

et al. 2018

View full text Add to dashboard Cite

During the 1980s Michie defined Machine Learning in terms of two orthogonal axes of performance: predictive accuracy and comprehensibility of generated hypotheses. Since predictive accuracy was readily measurable and comprehensibility not so, later definitions in the 1990s, such as Mitchell's, tended to use a one-dimensional approach to Machine Learning based solely on predictive accuracy, ultimately favouring statistical over symbolic Machine Learning approaches. In this paper we provide a definition of comprehensibility of hypotheses which can be estimated using human participant trials. We present two sets of experiments testing human comprehensibility of logic programs. In the first experiment we test human comprehensibility with and without predicate invention. Results indicate comprehensibility is affected not only by the complexity of the presented program but also by the existence of anonymous predicate symbols. In the second experiment we directly test whether any state-of-the-art ILP systems are ultra-strong learners in Michie's sense, and select the Metagol system for use in humans trials. Results show participants were not able to learn Editors: James Cussens and Alessandra Russo. Mach Learn (2018Learn ( ) 107:1119Learn ( -1140 the relational concept on their own from a set of examples but they were able to apply the relational definition provided by the ILP system correctly. This implies the existence of a class of relational concepts which are hard to acquire for humans, though easy to understand given an abstract explanation. We believe improved understanding of this class could have potential relevance to contexts involving human learning, teaching and verbal interaction.

show abstract

Section: Comprehensibility and Explanation In Machine Learningmentioning

confidence: 99%

Ultra-Strong Machine Learning: comprehensibility of programs learned with ILP

et al. 2018

View full text Add to dashboard Cite

show abstract

“…Let H 1 = u∈U 1 Hu denote the Hilbert space of functions that have additive form F (x) = u∈U 1 fu(xu) on univariate compnents; we call those components shape functions [19]. Similarly let H = u∈U Hu denote the Hilbert space of functions of x = (x1, ..., xn) that have additive form…”

Section: Problem Definitionmentioning

confidence: 99%

“…Standard additive modeling only involves modeling individual features (also called feature shaping). Previous research showed that gradient boosting with ensembles of shallow regression trees is the most accurate method among a number of alternatives [19].…”

Section: Fitting Generalized Additive Modelsmentioning

confidence: 99%

“…Previous research showed that feature shaping using gradient boosting [10] with shallow regression tree ensembles can achieve the best accuracy [19]. We follow similar approach (i.e., gradient boosting with shallow tree-like ensembles) in this work.…”

Section: Implementation Detailsmentioning

confidence: 99%

“…Standard GAMs are easy to interpret since users can visualize the relationship between the univariate terms of the GAM and the dependent variable through a plot fi(xi) vs. xi. However there is unfortunately a significant gap between the performance of the best standard GAMs and full complexity models [19]. In particular, Equation 1 does not model any interactions between features, and it is this limitation that lies at the core of the lack of accuracy of standard GAMs as compared to full complexity models.…”

mentioning

confidence: 99%

See 2 more Smart Citations

Accurate intelligible models with pairwise interactions

Lou

Caruana

Gehrke

et al. 2013

Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

Self Cite

328

285

View full text Add to dashboard Cite

Standard generalized additive models (GAMs) usually model the dependent variable as a sum of univariate models. Although previous studies have shown that standard GAMs can be interpreted by users, their accuracy is significantly less than more complex models that permit interactions.In this paper, we suggest adding selected terms of interacting pairs of features to standard GAMs. The resulting models, which we call GA 2 M-models, for Generalized Additive Models plus Interactions, consist of univariate terms and a small number of pairwise interaction terms. Since these models only include one-and two-dimensional components, the components of GA 2 M-models can be visualized and interpreted by users. To explore the huge (quadratic) number of pairs of features, we develop a novel, computationally efficient method called FAST for ranking all possible pairs of features as candidates for inclusion into the model.In a large-scale empirical study, we show the effectiveness of FAST in ranking candidate pairs of features. In addition, we show the surprising result that GA 2 M-models have almost the same performance as the best full-complexity models on a number of real datasets. Thus this paper postulates that for many problems, GA 2 M-models can yield models that are both intelligible and accurate.

show abstract

Asking ‘Why’ in AI: Explainability of intelligent systems – perspectives and challenges

Preece

2018

Intell Sys Acc Fin Mgmt

101

View full text Add to dashboard Cite

Summary Recent rapid progress in machine learning (ML), particularly so‐called ‘deep learning’, has led to a resurgence in interest in explainability of artificial intelligence (AI) systems, reviving an area of research dating back to the 1970s. The aim of this article is to view current issues concerning ML‐based AI systems from the perspective of classical AI, showing that the fundamental problems are far from new, and arguing that elements of that earlier work offer routes to making progress towards explainable AI today.

show abstract

Intelligible models for classification and regression

Cited by 370 publications

References 16 publications

Ultra-Strong Machine Learning: comprehensibility of programs learned with ILP

Ultra-Strong Machine Learning: comprehensibility of programs learned with ILP

Accurate intelligible models with pairwise interactions

Asking ‘Why’ in AI: Explainability of intelligent systems – perspectives and challenges

Contact Info

Product

Resources

About