Overview

Machine learning (ML) models are increasingly being employed to make highly consequential decisions pertaining to employment, bail, parole, and lending. While such models can learn from large amounts of data and are often very scalable, their applicability is limited by certain safety challenges. A key challenge is identifying and correcting systematic patterns of mistakes made by ML models before deploying them in the real world.

The goal of this workshop, held at the 2019 International Conference on Learning Representations (ICLR), is to bring together researchers and practitioners with different perspectives on debugging ML models.

Speakers

Aleksander Madry
MIT

Cynthia Rudin
Duke University

Dan Moldovan
Google

Deborah Raji
University of Toronto

Osbert Bastani
University of Pennsylvania

Sameer Singh
UC Irvine

Suchi Saria
Johns Hopkins University

Schedule

See here for a printable version.

Posters

Call for submissions (deadline has passed)

• Discovery of Intersectional Bias in Machine Learning Using Automatic Subgroup Generation. Angel Cabrera, Minsuk Kahng, Fred Hohman, Jamie Morgenstern and Duen Horng Chau
• Calibration of Encoder Decoder Models for Neural Machine Translation. Aviral Kumar and Sunita Sarawagi.
• Step-wise Sensitivity Analysis: Identifying Partially Distributed Representations for Interpretable Deep Learning. Botty Dimanov and Mateja Jamnik
• Handling Bias in AI Using Simulation. Daniel McDuff, Roger Cheng and Ashish Kapoor
• Inverting Layers of a Large Generator. David Bau, Jun-Yan Zhu, William Peebles, Hendrik Strobelt, Jonas Wulff, Bolei Zhou and Antonio Torralba
• MAST: A Tool for Visualizing CNN Model Architecture Searches. Dylan Cashman, Adam Perer and Hendrik Strobelt.
• Visualizations of Decision Regions in the Presence of Adversarial Examples. Grzegorz Swirszcz, Brendan O’Donoghue and Pushmeet Kohli.
• BertViz: A Tool for Visualizing Multi-Head Self-Attention in the BERT Model. Jesse Vig.
• Where To Be Adversarial Perturbations Added? Investigating and Manipulating Pixel Robustness Using Input Gradients. Jisung Hwang, Younghoon Kim, Sanghyuk Chun, Jaejun Yoo, Ji-Hoon Kim and Dongyoon Han.
• Dissecting Pruned Neural Networks. Jonathan Frankle and David Bau.
• Monitoring Opaque Learning Systems. Leilani Gilpin.
• Model Agnostic Globally Interpretable Explanations. Piyush Gupta, Nikaash Puri, Sukriti Verma, Pratiksha Agarwal and Balaji Krishnamurthy.
• Debugging Trained Machine Learning Models Using Flip Points. Roozbeh Yousefzadeh and Dianne O’Leary.
• Universal Multi-Party Poisoning Attacks. Saeed Mahloujifar, Ameer Mohammed and Mohammad Mahmoody.
• Failing Loudly: An Empirical Study of Methods for Detecting Dataset Shift. Stephan Rabanser, Stephan Guennemann and Zachary Lipton.
• Empirically Measuring Concentration: Fundamental Limits on Intrinsic Robustness. Xiao Zhang, Saeed Mahloujifar, Mohammad Mahmoody and David Evans.
• Taking a HINT: Leveraging Explanations to Make Vision and Language Models More Grounded. Ramprasaath R. Selvaraju, Stefan Lee, Yilin Shen, Hongxia Jin, Shalini Ghosh, Dhruv Batra and Devi Parikh.
• Similarity of Neural Network Representations Revisited. Simon Kornblith, Mohammad Norouzi, Honglak Lee and Geoffrey Hinton (Contributed talk).
• NeuralVerification.jl: Algorithms for Verifying Deep Neural Networks. Changliu Liu, Tomer Arnon, Christopher Lazarus and Mykel Kochenderfer (Contributed talk).
• Debugging Machine Learning via Model Assertions. Daniel Kang, Deepti Raghavan, Peter Bailis and Matei Zaharia (Contributed talk).
• The Scientific Method in the Science of Machine Learning. Jessica Zosa Forde and Michela Paganini (Contributed talk).

Demos

Call for submissions (deadline has passed)

• Operationalising Risk Management for Machine Learning: Building a Protocol-Driven System for Performance, Explainability, and Fairness. Imran Ahmed, Giles L. Colclough, Daniel First and QuantumBlack contributors.
• Building Models for Mobile Video Understanding. Franck Ngamkan and Geneviève Patterson.
• Debugging Large Scale Deep Recommender Systems using uncertainty estimations and attention. Inbar Naor, Ofer Alper, Dan Friedman and Gil Chamiel
• Adversarial Examples for Electrocardiograms. Xintian Han, Yuxuan Hu, Luca Foschini, Lior Jankelson and Rajesh Ranganath.
• Debuggable Machine Learning with ConX and Comet.ml. Cecelia Shao and Douglas Blank.
• Evidence Based Debugging with DRL-Monitor. Giang Dao and Minwoo Lee.
• Black Box Attacks on Transformer Language Models. Vedant Misra
• Improving Jobseeker-Employer Match Models at Indeed Through Process, Visualization, and Exploration. Benjamin Link, Eric Lawrence, Rosemarie Scott, Aaron Pigeon and Jon Witte (Contributed talk).
• MODHILL: A framework for debugging gait in multi-factor authentication systems. Vinay Prabhu, John Whaley and Mihail D.
• Who learns? A microscope into neural network training by measuring per-parameter learning. Janice Lan, Rosanne Liu, Hattie Zhou and Jason Yosinski.
• TensorWatch: A Multifaceted System for the Deep Learning Debugging and Visualization. Shital Shah, Roland Fernandez and Steven Drucker.
• Error terrain analysis for machine learning: Tool and visualizations. Rick Barraza, Russell Eames, Yan Esteve Balducci, Josh Hinds, Scott Hoogerwerf, Eric Horvitz, Ece Kamar, Jacquelyn Krones, Josh Lovejoy, Parham Mohadjer, Ben Noah and Besmira Nushi (Contributed talk).

Topics

• Debugging via interpretability: How can interpretable models and techniques aid us in effectively debugging ML models?

• Program verification as a tool for model debugging: Are existing program verification frameworks readily applicable to ML models? If not, what are the gaps that exist and how do we bridge them?

• Visualization tools for debugging ML models: What kind of visualization techniques would be most effective in exposing vulnerabilities of ML models?

• Human-in-the-loop techniques for model debugging: What are some of the effective strategies for using human input and expertise for debugging ML models?

• Novel adversarial attacks for highlighting errors in model behavior: How do we design adversarial attacks that highlight vulnerabilities in the functionality of ML models?

• Theoretical correctness of model debugging techniques: How do we provide guarantees on the correctness of proposed debugging approaches? Can we take cues from statistical considerations such as multiple testing and uncertainty to ensure that debugging methodologies and tools actually detect ‘true’ errors?

• Theoretical guarantees on the robustness of ML models: Given a ML model or system, how do we bound the probability of its failures?

• Insights into errors or biases of real-world ML systems: What can we learn from the failures of widely deployed ML systems? What can we say about debugging for different types of biases, including discrimination?

• Best practices for debugging large-scale ML systems: What are standardized best practices for debugging large-scale ML systems? What are existing tools, software, and hardware, and how might they be improved?

• Domain-specific nuances of debugging ML models in healthcare, criminal justice, public policy, education, and other social good applications.

See a list of references.

Organizers

Himabindu Lakkaraju
Harvard University

Sarah Tan
Cornell University / UCSF

Julius Adebayo
MIT

Jacob Steinhardt
Open Philanthropy Project / OpenAI

D Sculley
Google

Rich Caruana
Microsoft Research

Contact Us

Email debugging.ml@gmail.com any questions.

Sponsors

Program Committee