BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Xue-Cheng Tai (Hong Kong Baptist University) DTSTART:20211209T170000Z DTEND:20211209T180000Z DTSTAMP:20260423T035752Z UID:Inverse/66 DESCRIPTION:Title: Deep neural networks in image processing\nby Xue-Cheng Tai (Hong Kong Baptist University) as part of International Zoom Inverse Problems Semina r\, UC Irvine\n\n\nAbstract\nIn this talk\, we present our recent research on using variational models as layers for deep neural networks (DNNs). We use image segmentation as an example. The technique can also be used for high dimensional data classification as well. Through this technique\, we could integrate many well-know variational models for image segmentation i nto deep neural networks. The new networks will have the advantages of tra ditional DNNs. At the same time\, the outputs from the new networks can al so have many good properties of variational models for image segmentation. We will present some techniques to incorporate shape priors into the netw orks through the variational layers. We will show how to design networks w ith spatial regularization and volume preservation. We can also design net works with guarantee that the output shapes from the network for image seg mentation must be convex shapes/star-shapes. It is numerically verified th at these techniques can improve the performance when the true shapes satis fy these priors. \n\nThe ideas of these new networks is based on some rela tionship between the softmax function\, the Potts models and the structure of traditional DNNs. We will explain this in detail which leads naturally to the newly designed networks. \n\nThis talk is based on joint works wit h Jun Liu\, S. Luo and several other collaborators.\n LOCATION:https://researchseminars.org/talk/Inverse/66/ END:VEVENT END:VCALENDAR