List of research publications which were enabled by Modulos AutoML or to which Modulos team members contributed.

Image feature extraction and galaxy classification: a novel and efficient approach with automated machine learning

Federica Tarsitano, 
Claudio Bruderer
Kevin Schawinski
William G. Hartley, 
Submitted to the Monthly Notices of the Royal Astronomical Society
We extract features from galaxy images by computing the elliptical isophotes and projecting them into one-dimesional sequences. Then, we train and classify the sequences with Machine Learning (ML) algorithms using Modulos AutoML. We finally apply the best ML to the second public release of the Dark Energy Survey (DES DR2) data, a state-of-the-art astronomical imaging survey. We show we are able to successfully distinguish between early-type and late-type galaxies for images with signal-to-noise ratio greater then 300. This yields an accuracy on par with most contemporary automated image classification approaches.

Ease.ML: A Lifecycle Management System for MLDev and MLOps

Leonel Aguilar, 
David Dao, 
Shaoduo Gan, 
Nezihe Merve Gurel, 
Nora Hollenstein, 
Jiawei Jiang, 
Bojan Karlas
Thomas Lemmin, 
Tian Li, 
Yang Li, 
Susie Rao, 
Johannes Rausch, 
Cedric Renggli, 
Luka Rimanic, 
Maurice Weber, 
Shuai Zhang, 
Zhikuan Zhao, 
Kevin Schawinski
Wentao Wu, 
Ce Zhang
CIDR 2021
We present Ease.ML, a lifecycle management system for machine learning (ML). Unlike many existing works, which focus on im- proving individual steps during the lifecycle of ML application de- velopment, Ease.ML focuses on managing and automating the en- tire lifecycle itself.
A Lifecycle Management System for MLDev and MLOps

Continuous Integration of Machine Learning Models with Towards a Rigorous Yet Practical Treatment

Cedric Renggli, 
Bojan Karlaš
Bolin Ding, 
Feng Liu, 
Kevin Schawinski
Wentao Wu, 
Ce Zhang
Conference on Systems and Machine Learning (SysML) 2019
We present the first continuous integration system for machine learning. We design a domain specific language that allows users to specify integration conditions with reliability constraints, and develop simple novel optimizations that can lower the number of labels required by up to two orders of magnitude for test conditions popularly used in real production systems.
Integration of Machine Learning Models with Ease