基于对抗生成网络的多风格化的汉字

随着生成对抗网络（GAN）的发展,中文字体转换领域的研究越来越多,研究者能够生成高质量的汉字图像。这些字体转换模型可以使用GAN将源字体转换为目标字体。然而,目前的方法有以下局限:1）生成的图像模糊;2）模型一次只能学习和生成一种目标字体。针对这些问题,该文开发了一种全新的模式来执行中文字体转换。首先,将字体信息附加到图像上,告诉生成器需要转换的字体;然后,通过卷积网络提取和学习特征映射,并使用转置卷积网络生成照片真实图像。使用真实图像作为监控信息,以确保生成的字符和字体与它们自身一致。这个模型只需要训练一次,就能够将一种字体转换为多种字体并生成新的字体。对7个中文字体数据集的大量实验表明,该方法在中文字体转换中优于其他几种方法。     

PROSE: Proactive, Selective CDN Participation for P2P Streaming

Many production peer-to-peer (P2P) streaming systems use content delivery networks (CDN) to protect the user’s quality of experiences. Thus, how to efficiently utilize the capacity of CDN (e.g., which peers receive services from the CDN nodes) is a problem of practical significance. Existing solutions adopt a passive, on-demand approach, which is inefficient in utilizing CDN resources. In this paper, we propose PROSE, a simple, novel scheme to achieve proactive, selective CDN participation for P2P streaming. PROSE introduces novel concepts such as choke point expansion nodes/super nodes and leads to efficient, light-weighted, and distributed algorithms to identify and serve these nodes using CDN. Our experimental results show that PROSE achieves at least 10%～25% performance improvement and 2～4 times overhead reduction compared with existing general CDN-P2P-hybrid schemes.     

沙盐三井井下动态分析

本文分析了沙盐三井经过第四次修井后，卤水浓度有较大提高，产出大于产能及与新老溶腔的联系。