Distributionally Robust Deep Learning using Hardness Weighted Sampling

Lucas Fidon1, Michael Aertsen2, Thomas Deprest2, Doaa Emam3, Frédéric Guffens2, Nada Mufti4, Esther Van Elslander2, Ernst Schwartz5, Michael Ebner4, Daniela Prayer5, Gregor Kasprian5, Anna L David6, Andrew Melbourne4, Sébastien Ourselin4, Jan Deprest3, Georg Langs5, Tom Vercauteren4
1: Shool of Biomedical Engineering & Imaging Sciences, King's College London, 2: Department of Radiology, University Hospitals Leuven, 3: Department of Obstetrics and Gynaecology, University Hospitals Leuven, 4: School of Biomedical Engineering & Imaging Sciences, King’s College London, 5: Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna, 6: Institute for Women’s Health, University College London
PIPPI 2021 special issue
Publication date: 2022/07/18
PDF · arXiv · Video · Code

Abstract

Limiting failures of machine learning systems is of paramount importance for safety-critical applications. In order to improve the robustness of machine learning systems, Distributionally Robust Optimization (DRO) has been proposed as a generalization of Empirical Risk Minimization (ERM). However, its use in deep learning has been severely restricted due to the relative inefficiency of the optimizers available for DRO in comparison to the wide-spread variants of Stochastic Gradient Descent (SGD) optimizers for ERM.
We propose SGD with hardness weighted sampling, a principled and efficient optimization method for DRO in machine learning that is particularly suited in the context of deep learning. Similar to a hard example mining strategy in practice, the proposed algorithm is straightforward to implement and computationally as efficient as SGD-based optimizers used for deep learning, requiring minimal overhead computation. In contrast to typical ad hoc hard mining approaches, we prove the convergence of our DRO algorithm for over-parameterized deep learning networks with ReLU activation and finite number of layers and parameters.
Our experiments on fetal brain 3D MRI segmentation and brain tumor segmentation in MRI demonstrate the feasibility and the usefulness of our approach. Using our hardness weighted sampling for training a state-of-the-art deep learning pipeline leads to improved robustness to anatomical variabilities in automatic fetal brain 3D MRI segmentation using deep learning and to improved robustness to the image protocol variations in brain tumor segmentation.a decrease of 2% of the interquartile range of the Dice scores for the enhanced tumor and the tumor core regions.
Our code is available at https://github.com/LucasFidon/HardnessWeightedSampler

Keywords

Machine Learning · Image Segmentation · Distributionally Robust Optimization

Bibtex @article{melba:2022:019:fidon, title = "Distributionally Robust Deep Learning using Hardness Weighted Sampling", author = "Fidon, Lucas and Aertsen, Michael and Deprest, Thomas and Emam, Doaa and Guffens, Frédéric and Mufti, Nada and Van Elslander, Esther and Schwartz, Ernst and Ebner, Michael and Prayer, Daniela and Kasprian, Gregor and David, Anna L and Melbourne, Andrew and Ourselin, Sébastien and Deprest, Jan and Langs, Georg and Vercauteren, Tom", journal = "Machine Learning for Biomedical Imaging", volume = "1", issue = "PIPPI 2021 special issue", year = "2022", issn = "2766-905X", url = "https://melba-journal.org/papers/2022:019.html" }
RISTY - JOUR AU - Fidon, Lucas AU - Aertsen, Michael AU - Deprest, Thomas AU - Emam, Doaa AU - Guffens, Frédéric AU - Mufti, Nada AU - Van Elslander, Esther AU - Schwartz, Ernst AU - Ebner, Michael AU - Prayer, Daniela AU - Kasprian, Gregor AU - David, Anna L AU - Melbourne, Andrew AU - Ourselin, Sébastien AU - Deprest, Jan AU - Langs, Georg AU - Vercauteren, Tom PY - 2022 TI - Distributionally Robust Deep Learning using Hardness Weighted Sampling T2 - Machine Learning for Biomedical Imaging VL - 1 IS - PIPPI 2021 special issue SN - 2766-905X UR - https://www.melba-journal.org/papers/2022:019.html ER -

2022:019 cover