2017 Workshop on Human Interpretability in Machine Learning (WHI)

August 10, 2017

Sydney, Australia

click here for the 2018 edition of the workshop


The Second Annual Workshop on Human Interpretability in Machine Learning (WHI 2017), held in conjunction with ICML 2017, will bring together researchers who study the interpretability of predictive models, develop interpretable machine learning algorithms, and develop methodology to interpret black-box machine learning models. They will exchange ideas on these and allied topics, including:

  • Quantifying and axiomatizing interpretability,
  • Psychology of human concept learning,
  • Rule learning,
  • Symbolic regression,
  • Case-based reasoning,
  • Generalized additive models,
  • Interpretation of black-box models (including deep neural networks),
  • Causality of predictive models,
  • Visual analytics, and
  • Interpretability in reinforcement learning.

Location and Registration

The workshop will take place in Room C4.8 at the International Convention Centre, Sydney, Australia. Please consult the main ICML website for details on registration.

Workshop Proceedings

The accepted papers of the workshop may be found here: https://arxiv.org/html/1708.02666.


  • 8:30 A. Dhurandhar, V. Iyengar, R. Luss, and K. Shanmugam, "A Formal Framework to Characterize Interpretability of Procedures" [paper][presentation][reviews]
  • 8:45 A. Henelius, K. Puolamäki, and A. Ukkonen, "Interpreting Classifiers through Attribute Interactions in Datasets" [paper][presentation][reviews]
  • 9:00 S. Lundberg and S.-I. Lee, "Consistent Feature Attribution for Tree Ensembles" [paper][presentation][reviews]
  • 9:15 Invited Talk: D. Sontag [presentation]
  • 10:00 Coffee Break
  • 10:30 S. Penkov and S. Ramamoorthy, "Program Induction to Interpret Transition Systems" [paper][presentation][reviews] (best paper second runner-up)
  • 10:45 R. L. Phillips, K. H. Chang, and S. Friedler, "Interpretable Active Learning" [paper][presentation][reviews]
  • 11:00 C. Rosenbaum, T. Gao, and T. Klinger, "e-QRAQ: A Multi-turn Reasoning Dataset and Simulator with Explanations" [paper][presentation][reviews]
  • 11:15 Invited Talk: T. Jebara, "Interpretability Through Causality" [presentation]
  • 12:00 Lunch Break
  • 14:00 W. Tansey, J. Thomason, and J. G. Scott, "Interpretable Low-Dimensional Regression via Data-Adaptive Smoothing" [paper][presentation][reviews]
  • 14:15 Invited Talk: P. W. Koh [presentation]
  • 15:00 Coffee Break
  • 15:30 I. Valera, M. F. Pradier, and Z. Ghahramani, "General Latent Feature Modeling for Data Exploration Tasks" [paper][presentation][reviews] (best paper award)
  • 15:45 A. Weller, "Challenges for Transparency" [paper][presentation][reviews] (best paper runner-up)
  • 16:00 Awards Ceremony
  • 16:05 Panel Discussion: "Human Interpretability in Machine Learning" with panelists: Tony Jebara, Been Kim, Bernhard Schölkopf, and Kush Varshney; moderated by Adrian Weller

Invited Speakers

  • Tony Jebara, Columbia University and Netflix
    • Interpretability Through Causality
      • While interpretability often involves finding more parsimonious or sparser models to facilitate human understanding, Netflix also seeks to achieve human interpretability by pursuing causal learning. Predictive models can be impressively accurate in a passive setting but might disappoint a human user who expects the recovered relationships to be causal. More importantly, a predictive model's outcomes may no longer be accurate if the input variables are perturbed through an active intervention. I will briefly discuss applications at Netflix across messaging, marketing and originals promotion which leverage causal modeling in order to achieve models that can be actionable as well as interpretable. In particular, techniques such as two stage least squares (2SLS), instrumental variables (IV), extensions to generalized linear models (GLMs), and other causal methods will be summarized. These causal models can surprisingly recover more interpretable and simpler models than their purely predictive counterparts. Furthermore, sparsity can potentially emerge when causal models ignore spurious relationships that might otherwise be recovered in a purely predictive objective function. In general, causal models achieve better results algorithmically in active intervention settings and enjoy broader adoption from human stakeholders.
  • Pang Wei Koh, Stanford University
  • David Sontag, Massachusetts Institute of Technology

Previous Editions of the Workshop

Call for Papers and Submission Instructions

We invite submissions of full papers as well as works-in-progress, position papers, and papers describing open problems and challenges. While original contributions are preferred, we also invite submissions of high-quality work that has recently been published in other venues or is concurrently submitted.

Papers should be 4-6 pages in length (excluding references and acknowledgements) formatted using the ICML template and submitted online at this link. We expect submissions to be 4 pages but will allow up to 6 pages. The review process will be open and thus the submissions need not be anonymized.

Accepted papers will be selected for a short oral presentation or poster presentation.

Key Dates

  • Submission deadline: June 16, 2017
  • Notification: June 30, 2017
  • Workshop: August 10, 2017



We acknowledge generous support from