The What-If Tool: Interactive Probing of Machine Learning Models

By James Wexler, Mahima Pushkarna, Tolga Bolukbasi, Martin Wattenberg, Fernanda B. Viégas, Jimbo Wilson. 2019

600 citations recorded by the source corpus.

A key challenge in developing and deploying Machine Learning (ML) systems is understanding their performance across a wide range of inputs. To address this challenge, we created the What-If Tool, an open-source application that allows practitioners to probe, visualize, and analyze ML systems, with minimal coding. The What-If Tool lets practitioners test performance in hypothetical situations, analyze the importance of different data features, and visualize model behavior across multiple models and subsets of input data. It also lets practitioners measure systems according to multiple ML fairness metrics. We describe the design of the tool, and report on real-life usage at different organizations.

Subfields: Artificial Intelligence, Safety Research

How researchers see this paper

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

In Artificial Intelligence

In Safety Research

Citations in the corpus

ViCE: visual counterfactual explanations for machine learning models (2020, methodology): “WITshows howmodelpredictionschange after the inference of data, while our tool, visualizes how to inference from data in order to change a prediction into other classes.”
Explainable machine learning in deployment (2019, background): “ist [65, 68], they either work for specific model-types or are not black-box in nature.”
A survey of algorithmic recourse: definitions, formulations, solutions, and prospects (2020, methodology): “E grad opt gradient (2019.05) CERTIFAI [162] E gen alg query (2019.06) MACEM [53] E FISTA query (2019.06) Expl. using SHAP [152] E heuristic query (2019.07) Nearest Observable [181] E brute force dataset (2019.07) Guided Prototypes [173] E grad opt/FISTA gradient/query (2019.07) REVISE [87] E grad o”
The Language Interpretability Tool: Extensible, Interactive Visualizations and Analysis for NLP Models (2020, background): “Many are general-purpose, such as the What-If Tool (Wexler et al., 2020), Captum (Kokhlikyan et al.”
Multi-Objective Counterfactual Explanations (2020, neutral): “. As a baseline, we selected the closest observed data point to x (according to the Gower distance) that has a prediction equal to our desired outcome. Since this approach is part of the What-If Tool [36], we call this approach ‘Whatif’. The parameters of DiCE, Recourse and Tweaking were set to the ”
Counterfactual Explanations for Machine Learning: A Review (2020, background): “Wexler et al. [112] developed a model-agnostic interactive visual tool for letting developers and practitioners visually examine the effect of changes in various features.”
Understanding the Effect of Out-of-distribution Examples and Interactive Explanations on Human-AI Decision Making (2021, background): “Specifically, interactive machine learning tends to involve machine learning practitioners, while our work considers the population that does not have a machine learning background [24, 31, 61, 67].”
Interpretable Machine Learning - A Brief History, State-of-the-Art and Challenges (2020, background): “IML has also arrived in industry [36], there are startups that focus on ML interpretability and also big tech companies offer software [126,8,43].”
Are We Learning Yet? A Meta Review of Evaluation Failures Across Machine Learning (2021, background): “…to the benchmarking paradigm, such as test suites [1, 126, 170], for example, can re-orient ML evaluation away from its current focus on the competitive determination of “state of the art”, and more towards an exploratory and descriptive probing of model capabilities [65, 106, 143, 161, 169].”
Algorithmic Recourse: from Counterfactual Explanations to Interventions (2020, background): “A counterfactual instance can be from the dataset (Wexler et al., 2019; Poyiadzi et al., 2019), or generated as in (Wachter et al.”
Robustness Gym: Unifying the NLP Evaluation Landscape (2021, neutral): “, 2020), What-if (Wexler et al., 2019), FairVis (Cabrera et al.”
Counterfactual Explanations and Algorithmic Recourses for Machine Learning: A Review (2020, neutral): “[333] have developed interactive graphical user interfaces for displaying CFEs.”