首页科学研究学术预告 正文

Efficient Neural Architecture Search

【 发布日期:2023-04-07 】    作者:罗昕

报告题目:Efficient Neural Architecture Search


报告人:常晓军博士澳大利亚悉尼科技大学教授澳大利亚人工智能研究所ReLER实验室主任


报告时间:20234714


报告地点:山东大学软件园校区办公楼202会议室


报告摘要:Neural Architecture Search (NAS) has emerged as a promising approach to automatically design neural network architectures that can achieve state-of-the-art performance on various tasks. However, traditional NAS methods are computationally expensive and require a large amount of resources, hindering their practical use in many real-world applications. Efficient Neural Architecture Search aims to address these issues by developing methods that can efficiently explore the architecture space and find high-performing architectures using fewer resources. This talk will provide an overview of different approaches to Efficient NAS, including methods that leverage transfer learning, network pruning, and weight sharing to reduce the search space and accelerate the search process. We will also discuss recent advances in Efficient NAS, such as methods that incorporate reinforcement learning and meta-learning techniques to further improve the efficiency of the search process. Finally, we will highlight some of the practical applications of Efficient NAS, including designing neural networks for mobile and edge devices, as well as optimizing the architecture of large-scale deep learning models used in computer vision, natural language processing, and other domains.


报告人简介:常晓军博士目前为澳大利亚悉尼科技大学教授、澳大利亚人工智能研究所ReLER实验室主任。悉尼科技大学在澳大利亚计算机方向名列前茅,在2022 USNews全球大学排名榜单中,计算机科学位列全球第13位。常教授主要从事人工智能、计算机视觉和多媒体技术等方向的研究,在无约束视频、监控视频复杂事件检测和高效深度学习等方面的研究具有特色和优势,取得了多项具有国际影响力的重要创新性成果。基于突出的学术成果、创新成果,常教授于2018年荣获澳大利亚研究委员会青年研究奖 (ARC DECRA)。近五年来发表高水平论文共149篇,包括模式识别和计算机视觉领域顶级期刊IEEE T-PAMI、计算机视觉领域顶级期刊IJCV和领域内ACM/IEEE Transactions等国际顶级期刊论文73篇;以及相关领域顶级会议ICMLCVPRICCVICLRCCF A类论文64篇。上述论文中,常教授以第一作者和通讯作者发表论文32篇。常教授在复杂事件检测、零样本目标检测和高效深度学习等方面的研究成果已在相关领域内引起同行的广泛关注,发表论文在Google Scholar上的总引用一万余次;其中十四篇为ESI高被引论文。常教授在视频事件检测中原创性的成果及其发表的文章多次被国际著名科研团队在他们发表的国际顶级期刊和会议文章中引用,包括美国卡内基梅隆大学、哥伦比亚大学、UIUC以及Google BrainDeepMind的科研团队。常教授目前指导博士研究生十二名,其指导的学生获得了2020年澳大利亚Research Training Program (RTP)奖学金(澳大利亚最高级别国际学生奖学金),毕业的学生加入世界顶级名校美国卡内基梅隆大学、普林斯顿大学、英国牛津大学等从事博士后研究工作,或者加入国内顶尖985高校(包括上海交通大学、西安交通大学、北京航空航天大学等)担任Tenure Track助理教授。