報(bào)告題目：Learning Neural Networks for Domains with Few Labels
報(bào)告人：Joost van de Weijer教授
主持人：劉準(zhǔn)釓教授
報(bào)告時(shí)間：2019年11月04日09:30AM
報(bào)告地點(diǎn)：長(zhǎng)安校區(qū)自動(dòng)化學(xué)院341會(huì)議室
報(bào)告簡(jiǎn)介：Machine learning algorithms are data hungry and require a lot of labeled data to be trained. Exploitation of unlabeled data during training is thus a long-pursued objective of machine learning. In this talk, I will explore several methods which use unlabeled data to train deep neural networks. First, I will show how encoder-decoder architectures can be exploited to transfer labels from one domain to another (unlabeled) domain. Next, I will focus on how rankings can be used as a self-supervised proxy task to include unlabeled data during the training of deep neural networks. I will show applications in image quality assessment, crowd counting, and multi-modal data processing.
報(bào)告人簡(jiǎn)歷
Dr. Joost van de Weijer is the leader of the Learning and Machine Perception Team (LAMP) and a senior scientist at the Computer Vision Center in Barcelona. He received his MSc (Delft University of Technology) in 1998 and his PhD (University of Amsterdam) in 2005. He was Marie Curie Intra-European fellow in INRIA Rhone-Alpes. He was awarded the Ramon y Cajal Research Fellowship. His research interest includes deep learning, computer vision, color imaging, object recognition, lifelong learning, active learning, color image processing. He has many publications in the top conferences CVPR, ECCV, ICCV, NIPS and the major journals PAMI, IJCV, and TIP.
