Privacy Implications of Visual Data Dissemination (A02)

今天的用户通常会分享大量的视觉数据（图像、视频）。我们的长期目标是提供全面而准确的工具来评估所涉及的隐私威胁：可以提取用户的活动吗？用户可以在不同的帖子中重新识别吗？社会关系可以重建吗？我们将推进最先进的计算机视觉技术及其与在线上下文信息的结合，以及研究面部模糊等保护技术的效果。我们将设计概率模型来分析从一组图像中可以做出的推断，并预测在添加额外图像的情况下可以做出的推断（假设）。

Users today typically share massive amounts of visual data (images, videos). Our long-term goal is to provide comprehensive and accurate tools for assessing the privacy threats entailed: Can a users activities be extracted? Can a user be re-identified across separate postings? Can social relations be reconstructed? We will advance state-of-the-art computer vision techniques and their combination with online context information, as well as examine the effect of protection techniques such as face blurring. We will devise probabilistic models for analyzing the inferences that can be made from a set of images, and predict the inferences that could be made (what-if) in case additional images were added.