2017/06/03 by Philip Häusser, Alexander Mordvintsev, Häusser, Philip +3 · 3 citations
Computer Science · #Advanced Neural Network Applications #Adversarial Robustness in Machine Learning #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning and Data Classification
paper · pdf · doi:10.48550/arxiv.1706.00909
openalex publication_date 2017/06/03 · openalex created_date 2022/10/02 · openalex updated_date 2026/07/28
In many real-world scenarios, labeled data for a specific machine learning\ntask is costly to obtain. Semi-supervised training methods make use of\nabundantly available unlabeled data and a smaller number of labeled examples.\nWe propose a new framework for semi-supervised training of deep neural networks\ninspired by learning in humans. "Associations" are made from embeddings of\nlabeled samples to those of unlabeled ones and back. The optimization schedule\nencourages correct association cycles that end up at the same class from which\nthe association was started and penalizes wrong associations ending at a\ndifferent class. The implementation is easy to use and can be added to any\nexisting end-to-end training setup. We demonstrate the capabilities of learning\nby association on several data sets and show that it can improve performance on\nclassification tasks tremendously by making use of additionally available\nunlabeled data. In particular, for cases with few labeled data, our training\nscheme outperforms the current state of the art on SVHN.\n