学术报告
报告题目:Communication Efficient Federated Learning
报 告 人:金耀初教授
报告时间:2020年10月2日(周五)下午16:00
腾讯会议:393 732 875
会议密码:123456
会议直播:https://meeting.tencent.com/l/X4dseQAD7ESz
主办单位:地下空间智能控制教育部工程研究中心,williamhill威廉希尔官网
报告摘要:
Federated learning is a new distributed learning paradigm that can preserve data privacy in machine learning. One of the main challenges in federated learning is to reduce the communication costs for transmitting model parameters between local devices and the central server and vice versa. This talk presents some most recent work on communication efficient federated learning, including constructing compact local models, introducing layer-wise asynchronous parameter update, and using ternary quantization. At the end of the talk, future directions of research in federated learning will be briefly discussed.
报告人简介:
金耀初,芬兰国家技术局“芬兰杰出教授”,全球高被引科学家,IEEE Fellow。
研究方向涉及人工智能的多个领域,包括数据驱动代理模型辅助的进化优化、可靠机器学习、多目标进化学习、集群机器人和进化发展系统。目前是IEEE Transactions on Cognitive and Developmental Systems和Complex & Intelligent Systems期刊主编。荣获2018和2020 IEEE Transactions on Evolutionary Computation最佳论文奖,2014、2016以及2019年IEEE Computational Intelligence Magazine最佳论文奖。