2017/03/04 by Markus Wulfmeier, Wulfmeier, Markus, Alex Bewley +3 · 3 citations
Computer Science · #Adversarial Robustness in Machine Learning #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Human Pose and Action Recognition #Machine Learning (cs.LG) #Robotics (cs.RO)
paper · pdf · doi:10.48550/arxiv.1703.01461
openalex publication_date 2017/03/04 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Appearance changes due to weather and seasonal conditions represent a strong\nimpediment to the robust implementation of machine learning systems in outdoor\nrobotics. While supervised learning optimises a model for the training domain,\nit will deliver degraded performance in application domains that underlie\ndistributional shifts caused by these changes. Traditionally, this problem has\nbeen addressed via the collection of labelled data in multiple domains or by\nimposing priors on the type of shift between both domains. We frame the problem\nin the context of unsupervised domain adaptation and develop a framework for\napplying adversarial techniques to adapt popular, state-of-the-art network\narchitectures with the additional objective to align features across domains.\nMoreover, as adversarial training is notoriously unstable, we first perform an\nextensive ablation study, adapting many techniques known to stabilise\ngenerative adversarial networks, and evaluate on a surrogate classification\ntask with the same appearance change. The distilled insights are applied to the\nproblem of free-space segmentation for motion planning in autonomous driving.\n