Making machine learning models interpretable

By Alfredo Vellido, José David Martín-Guerrero, Paulo J. G. Lisboa. 2012

316 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

Sanity Checks for Saliency Maps (2018, background): “As machine learning grows in complexity and impact, much hope rests on explanation methods as tools to elucidate important aspects of learned models [1, 2].”
Local Explanation Methods for Deep Neural Networks Lack Sensitivity to Parameter Values (2018, background): “Understanding how a trained model derives its output, as well as the factors responsible, is a central challenge in machine learning (Vellido et al., 2012; Doshi-Velez et al., 2017).”
Interventions over Predictions: Reframing the Ethical Debate for Actuarial Risk Assessment (2017, background): “In addition to transparency issues related to proprietary software, a major focus of recent scholarship has been on developing interpretable machine learning models that enable outside researchers to better scrutinize the underlying logic of actuarial tools which are based on machine learning (Zeng ”
Interpretable Two-level Boolean Rule Learning for Classification (2015, methodology): “As an example, a Boolean rule in [8] for the prediction of 10 year coronary heart disease (CHD) risk for a 45 year old male can be expressed as follows: IF 1.”
Learning customized and optimized lists of rules with mathematical programming (2018, background): “The goal of designing interpretable classifiers has existed since the beginning of artificial intelligence, (see reviews of [21,31,52,59,68]).”
The ethics of algorithms: Mapping the debate (2016, neutral): “…2016), using ex post audit studies (Adler et al., 2016; Diakopoulos, 2015; Kitchin, 2016; Romei and Ruggieri, 2014; Sandvig et al., 2014), reﬂexive ethnographic studies in development and testing (Neyland, 2016), or reporting mechanisms designed into the algorithm itself (Vellido et al., 2012).”
Machine learning, social learning and the governance of self-driving cars (2018, background): “2 For the machine-learning community, the challenge is seen as one of ‘interpretabil-ity’ (Vellido et al., 2012).”
Supersparse Linear Integer Models for Interpretable Classification (2013, background): “On the other hand, the statistics community has developed black-box classification models that are optimized for predictive accuracy but have mostly ignored interpretability (with a few exceptions, see [71]).”
Clinical Information Extraction Applications: A Literature Review (2017, background): “Other reasons include the challenge of interpretability of machine learning methods [174] and the difficulty of correcting specific errors reported by end users (compared to rule-based approaches, which can trivially modify rules correct specific errors).”
A Workflow for Visual Diagnostics of Binary Classifiers using Instance-Level Explanations (2017, background): “ons. While common practice in much of the machine learning endeavors is to focus on model accuracy, many researchers have voiced the need for more transparency when the application domain requires it [4,8,12,22,23,31]. A recent DARPA (Defense Advanced Research Projects Agency) program called “Explai”
Interpreting Black-Box Classifiers Using Instance-Level Visual Explanations (2017, background): “, healthcare, justice, security), verifying the model exclusively through the lens of statistical accuracy is often not su cient [4, 6, 8, 15, 16, 24].”
Visual Interaction with Deep Learning Models through Collaborative Semantic Inference (2019, background): “Interactively building this intuition is crucial to end-users since past research has shown that statically presenting only a few instances may lead to misrepresentation of the model limitations [44] or the data that the model uses [77].”