题目:针对分布式矩阵乘法的折叠多项式码
报告人:徐敬可
时间:12月17上午10点
地点:腾讯会议341 448 710
报告人简介:
徐敬可,2019年博士毕业于中国科学院数学与系统科学研究院,2020年10月起就职于新葡的京集团350vip8888新葡的京集团350vip8888,讲师。主要研究方向为分布式存储编码、保密信息提取以及分布式计算等。在IEEE Transactions on Information Theory、IEEE Transactions on Communications、SCIENCE CHINA Information Sciences以及ISIT等信息论与编码理论期刊与会议发表论文近十篇。主持省自然科学青年基金1项,国家自然科学青年基金1项,作为学术骨干成员参与国家自然科学面上基金1项,山东省高等学校青创科技计划创新团队核心成员。
报告简介:
In this talk, we consider the large scale matrix multiplication problem of computing $AA^\top$, which is in a distributed system comprising a group of worker nodes and a master node. For effective straggler mitigation, we propose a novel computation strategy, named folded polynomial code. Moreover, we demonstrate that folded code achieves the optimal recovery threshold among all linear computation strategies. Compared with MatDot codes, folded polynomial codes reduce the recovery threshold by factor of 2, and at least reduce the computational complexity in decoding process by factor of 2.