报告题目：Correlation Alignment for Domain Adaptation
报告摘要：Unlike human learning, machine learning often fails to handle changes between training (source) and test (target) input distributions. Such domain shifts, common in practical scenarios, severely damage the performance of conventional machine learning methods. Supervised domain adaptation methods have been proposed for the case when the target data have labels, including some that perform very well despite being “frustratingly easy” to implement. However, in practice, the target domain is often unlabeled, requiring unsupervised adaptation.
In this talk, I'll first present a simple, effective, and efficient method for unsupervised domain adaptation called CORrelation ALignment (CORAL). CORAL minimizes domain shift by aligning the second-order statistics of source and target distributions, without requiring any target labels. Even though it is extraordinarily simple, CORAL performs remarkably well in extensive evaluations on standard benchmark datasets.
In the second part of the talk, I'll talk about extensions that apply the CORAL framework to three different scenarios. For linear transformation, we equivalently applied CORAL to the classifier parameters, resulting in added efficiency when the number of classifiers is small but the number and dimensionality of target examples are very high. The resulting CORAL Linear Discriminant Analysis (CORAL-LDA) outperforms standard LDA by a large margin on standard domain adaptation benchmarks. The second scenario is applying CORAL to subspace manifold based methods. We incorporate CORAL to two recently published methods and the resulting CORAL-SS method outperforms its counterparts consistently. Last but not least, we extend CORAL to learn a nonlinear transformation that aligns correlations of layer activations in deep neural networks. Deep CORAL works seamlessly with deep networks and achieves state-of-the-art performance on standard benchmark datasets.
The last part of the talk explores two new case studies for domain adaptation: from virtual to reality and from ground to sky. In from virtual to reality, the source domain contains synthetic images generated from 3D CAD models while the target data are real images. For from ground to sky, the target domain contains aerial images while the source data are the common consumer photos.
报告人简介：Dr. Baochen Sun is currently a Senior Researcher at Microsoft working on Computer Vision and Deep Learning. He got his PhD in 2016 and joined Microsoft right after graduation. His PhD research includes domain adaptation, deep learning, computer vision, and machine learning. His work on domain adaptation has won one best paper award and one honorable mention paper. He is also the program chair of NIPS 2015 workshop on Transfer and Multi-Task Learning: Trends and New Perspectives. He is a reviewer/program committee of leading conferences/journals in Computer Vision and Machine Learning including ICML, NeurIPS, ICCV, CVPR, ECCV, ICLR, IJCAI, AAAI, TPAMI, IJCV, TIP, etc.