以图像视频为中心的跨媒体分析与推理

Image video centered cross-media analysis and reasoning

如何跨越从跨媒体数据到跨媒体知识所面临的“异构鸿沟”和“语义鸿沟”,对体量巨大的跨媒体数据进行有效管理与利用,是发展新一代人工智能亟待突破的瓶颈问题。针对以图像视频为代表的海量网络跨媒体内容,借鉴人类感知与认知机理,本文对跨媒体内容统一表征与符号化表征、跨媒体深度关联理解、类人跨媒体智能推理等关键技术开展研究。基于上述关键技术,着力于解决发展新一代人工智能的知识匮乏共性难题,开展大规模跨媒体知识图谱的构建及人机协同标注技术研究,为跨媒体感知进阶到认知提供关键支撑,进一步为跨媒体理解、检索、内容转换生成等跨媒体内容管理与服务热点应用领域提供了可行思路。

How to surpass the heterogeneity gap and semantic gap between the cross-media content and cross-media knowledge,and how to manage and utilize the huge amount of cross-media data effectively are urgent bottleneck prob-lems of developing a new generation of artificial intelligence.Aiming at massive online cross-media content represen-ted by image video and by referring to human perception and cognition mechanisms,this paper undertakes studies on such key technologies as unified representation and symbolic representation of cross-media content,deep correlative un-derstanding of cross-media and human-like cross-media intelligent reasoning.Based on the above technologies,this pa-per focuses on solving the common problem of knowledge shortage in the development of a new generation of artificial intelligence and carries out a research on the construction of large-scale cross-media knowledge graph and the human-machine cooperation based labeling technology,to provide strong support for the advancement from cross-media per-ception to cognition and further provide feasible solutions towards cross-media content management and popular ser-vice applications,e.g.,cross-media content understanding,retrieval,content transformation and generation,etc.