Journal of Information Security Research ›› 2019, Vol. 5 ›› Issue (11): 961-965.

Previous Articles     Next Articles

AI and Data Privacy Protection: The Way to Federated Learning


  • Received:2019-11-08 Online:2019-11-15 Published:2019-11-20



  1. 微众银行
  • 通讯作者: 杨强
  • 作者简介:杨强,博士,教授,香港科技大学新明工程学讲席教授,微众银行首席人工智能官.国际人工智能界“迁移学习”(transfer learning)技术的开创者,并提出“联邦学习”(federated learning)的研究新方向.于2013年7月当选为国际人工智能协会(AAAI)院士,是第1位获此殊荣的华人.于2016年5月当选为AAAI执行委员会委员,是首位也是至今为止唯一的AAAI华人执委.2017年8月当选为国际人工智能联合会(IJCAI)理事会主席,是第1位担任IJCAI理事会主席的华人科学家.主要研究方向为迁移学习、人工智能、大数据.

Abstract: With the tremendous advance in computing, algorithms and data volume, artificial intelligence ushered in the third development climax, and began to gain a foot hold in exploring various industries. However, as the emergence of “big data”, more “small data” or “poorquality data”, and “data silos” exist in industry applications. For example, in the information security realm, it is difficult for enterprises who provide security services such as content security auditing and intrusion detection based on artificial intelligence technology to exchange raw data due to the consideration of user privacy and trade secrets protection. The services between enterprises are independent, and the overall development of cooperation and technology is difficult to make a breakthrough in a short period of time. How to promote greater cooperation on the premise of protecting the privacy of organizations? Will there be any chance for technical means to solve the data privacy protection problems? Federated Learning is an effective way to solve this problem and achieve acrossenterprise collaborative governance.

Key words: artificial intelligence, federated learning, data security, data privacy, corporate collaborative governance

摘要: 伴随着计算力、算法和数据量的巨大进步,人工智能迎来第3次发展高潮,开始了各行业的落地探索.然而,在“大数据”兴起的同时,更多行业应用领域中是“小数据”或者质量很差的数据,“数据孤岛”现象广泛存在.例如在信息安全领域的应用中,虽然多家企业推出了基于人工智能技术的内容安全审核、入侵检测等安全服务,但出于用户隐私和商业机密的考虑,企业之间很难进行原始数据的交换,各个企业之间服务是独立的,整体协作和技术水平很难在短时间内实现突破式发展.如何在保护各机构数据隐私的前提下促成更大范围的合作,能否通过技术手段破解数据隐私保护难题,联邦学习是解决这一问题、实现跨企业协同治理的有效方式.

关键词: 人工智能, 联邦学习, 数据安全, 数据隐私, 企业协同治理