2016年12月6日杜克大学Prof. Lawrence Carin来访并作报告

报告人：Prof. Lawrence Carin, Duke University

报告题目：Variational Autoencoder for Deep Learning of Images, Labels and  Captions

报告摘要：A novel variational autoencoder is developed to model images, as well as  associated labels or captions. The Deep Generative Deconvolutional Network  (DGDN) is used as a decoder of the latent image features, and a deep  Convolutional Neural Network (CNN) is used as an image encoder; the CNN is used  to approximate a distribution for the latent DGDN features/code. The latent code  is also linked to generative models for labels (Bayesian support vector machine)  or captions (recurrent neural network). When predicting a label/caption for a  new image at test, averaging is performed across the distribution of latent  codes; this is computationally efficient as a consequence of the learned  CNN-based encoder. Since the framework is capable of modeling the image in the  presence/absence of associated labels/captions, a new semi-supervised setting is  manifested for CNN learning with images; the framework even allows unsupervised  CNN learning, based on images alone.