2018/11/17 by Clemens-Alexander Brust, Brust, Clemens-Alexander, Joachim Denzler +1 · 3 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Machine Learning in Bioinformatics
paper · pdf · doi:10.48550/arxiv.1811.07125
openalex publication_date 2018/11/17 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
One of the most prominent problems in machine learning in the age of deep\nlearning is the availability of sufficiently large annotated datasets. For\nspecific domains, e.g. animal species, a long-tail distribution means that some\nclasses are observed and annotated insufficiently. Additional labels can be\nprohibitively expensive, e.g. because domain experts need to be involved.\nHowever, there is more information available that is to the best of our\nknowledge not exploited accordingly. In this paper, we propose to make use of\npreexisting class hierarchies like WordNet to integrate additional domain\nknowledge into classification. We encode the properties of such a class\nhierarchy into a probabilistic model. From there, we derive a novel label\nencoding and a corresponding loss function. On the ImageNet and NABirds\ndatasets our method offers a relative improvement of 10.4% and 9.6% in accuracy\nover the baseline respectively. After less than a third of training time, it is\nalready able to match the baseline's fine-grained recognition performance. Both\nresults show that our suggested method is efficient and effective.\n