Mon 13 Jun 2022 16:30 - 16:45 at Boardroom - Evening Chair(s): Swarat Chaudhuri
Tue 14 Jun 2022 04:30 - 04:45 at Boardroom - Evening

Machine learning in practice often involves complex pipelines for data cleansing, feature engineering, preprocessing, and prediction. These pipelines are composed of operators, which have to be correctly connected and whose hyperparameters must be correctly configured. Unfortunately, it is quite common for certain combinations of datasets, operators, or hyperparameters to cause failures. Diagnosing and fixing those failures is tedious and error-prone and can seriously derail a data scientist’s workflow. This paper describes an approach for automatically debugging an ML pipeline, explaining the failures, and producing a remediation. We implemented our approach, which builds on a combination of AutoML and SMT, in a tool called Maro. Maro works seamlessly with the familiar data science ecosystem including Python, Jupyter notebooks, scikit-learn, and AutoML tools such as Hyperopt. We empirically evaluate our tool and find that for most cases, a single remediation automatically fixes errors, produces no additional faults, and does not significantly impact optimal accuracy nor time to convergence.

Mon 13 Jun

Displayed time zone: Pacific Time (US & Canada) change

15:30 - 17:00
EveningMAPS at Boardroom +12h
Chair(s): Swarat Chaudhuri University of Texas at Austin
15:30
45m
Keynote
Unsupervised Program Synthesis: Hierarchy and Perception
MAPS
Kevin Ellis Cornell University
16:15
15m
Talk
ExeBench: An ML-scale dataset of executable C functions
MAPS
Jordi Armengol-Estapé University of Edinburgh, Jackson Woodruff University of Edinburgh, Alexander Brauckmann University of Edinburgh, José Wesley de Souza Magalhães University of Edinburgh, Michael F. P. O'Boyle University of Edinburgh
16:30
15m
Talk
Automatically Debugging AutoML Pipelines Using Maro: ML Automated Remediation Oracle
MAPS
Julian Dolby IBM Research, USA, Jason Tsay IBM Research, Martin Hirzel IBM Research
16:45
15m
Talk
A Graph Neural Network-based performance model for Deep Learning Applications
MAPS
Shikhar Singh University of Texas, James Hegarty Facebook, Hugh Leather University of Edinburgh, UK, Benoit Steiner Facebook

Tue 14 Jun

Displayed time zone: Pacific Time (US & Canada) change

03:30 - 05:00
EveningMAPS at Boardroom
03:30
45m
Keynote
Unsupervised Program Synthesis: Hierarchy and Perception
MAPS
Kevin Ellis Cornell University
04:15
15m
Talk
ExeBench: An ML-scale dataset of executable C functions
MAPS
Jordi Armengol-Estapé University of Edinburgh, Jackson Woodruff University of Edinburgh, Alexander Brauckmann University of Edinburgh, José Wesley de Souza Magalhães University of Edinburgh, Michael F. P. O'Boyle University of Edinburgh
04:30
15m
Talk
Automatically Debugging AutoML Pipelines Using Maro: ML Automated Remediation Oracle
MAPS
Julian Dolby IBM Research, USA, Jason Tsay IBM Research, Martin Hirzel IBM Research
04:45
15m
Talk
A Graph Neural Network-based performance model for Deep Learning Applications
MAPS
Shikhar Singh University of Texas, James Hegarty Facebook, Hugh Leather University of Edinburgh, UK, Benoit Steiner Facebook