|9:45-10:30||Scalable and Learnable Evolutionary Multiobjective Optimization||刘松柏|
|10:30-11:15||Architectural Support for Trustworthy Computing on RISC-V||李拓|
我们提出一个全新的树突状神经元模型(Dendritic neural model)。该模型可以模拟大脑神经元的树突可塑性现象，自适应地裁剪冗余的突触和树突分支，来精简自身模型结构。对比于其他人工神经网络，该模型有着更高生物仿真性。同时，精简后的神经元模型可以编译成由比较器，逻辑与，或，非门构建的逻辑电路分类器(Logic circuit classifier)。对比于其他以浮点数计算为主的机器学习算法，该分类器完全以二进制计算，因此有着更快的运算速度和更低的计算功耗。考虑到其可以在FPGS和VLSI等上硬件实现和进行大规模并行计算，该分类器在处理大规模高数据流问题上有着更加明显的性能优势。也为有效解决“冯诺依曼瓶颈”提出全新的思路。
吉君恺，深圳大学新葡的京集团350vip8888网络安全研究所副研究员。研究兴趣包括神经网络，进化计算和计算机辅助药物设计。于2103年在合肥工业大学获得学士学位，于2016年和2018年在日本富山大学获得硕士和博士学位。累计发表学术论文40余篇，包括IEEE Transactions on Neural Networks and Learning Systems、IEEE Computational Intelligence Magazine等。主持国家自然科学基金青年项目和广东省区域联合基金。获得过2016年IEEE PIC国际会议最佳论文奖。
Scalable and Learnable Evolutionary M bjective Optimization Evolutionary algorithms characteriz
ultioed with a population-based iterative search engine have been recognized as an effective tool for solving multiobjective optimization problems in science and engineering. To continue being useful to society, evolutionary multiobjective optimization (EMO) has to address new challenges brought by complex optimization problems with continuously increasing scale (e.g., many-objective functions, large-scale decision variables, and multitask). This requires from EMO researchers new thinking and leads to the increasing demand of designing new EMO solvers for optimizing these complex optimization problems effectively. Since the number of nondominated solutions grows exponentially with the number of objective functions and the search space expands exponentially with the number of variables, learning to customize efficient environmental selection strategies in the objective space and simplify the search space or enhance the search ability in the variable space are crucial for improving the scalability of the designed EMO frameworks. Besides, as real-world optimization problems seldom exist in isolation, the experience of solving one problem (or task) may learn useful knowledge to assist the optimizing of other related ones. This is consistent with the learning behaviors of human beings that useful knowledge from past experiences can be exploited to solve relevant problems at hand and any potential synergies will be excavated even facing multiple seemingly unrelated problems. Thus, it is an attractive idea that knowledge can be transferred across related complex optimization problems to assist their optimization. This report focuses on the exploration and elucidation of our research work on scalable and learnable EMO related to the above three emerging hot topics: many-objective optimization, large-scale optimization, and multitask optimization.
Songbai Liu received the B.S. degree from Changsha University and the M.S. degree from Shenzhen University, China, in 2012 and 2018, respectively. He worked for ShenZhen TVT Digital Technology Co., Ltd as a software engineer from 2013 to 2015, and he worked for Shenzhen University as a research assistance from 2018 to 2019.
He is currently a Ph.D. candidate in Department of Computer Science, City University of Hong Kong, Hong Kong. He has published 10 papers on IEEE TEVC, IEEE TCYB, IEEE TSMC, Swarm EC, and Information Sciences. He is serving as a reviewer of over five international journals, including IEEE TEVC, IEEE TCYB, IEEE TSMC, MC, etc. He is a member of the IEEE Computational Intelligence Society. His current research interests include neural and evolutionary computing, evolutionary transfer optimization, and evolutionary large-scale optimization.
Architectural Support for Trustworthy Computing on RISC-V
Timing channel attack and memory safety corruption are critical security issues in modern computer systems. Timing channel has been used as one key component in notorious Meltdown and Spectre attacks, while memory safety violations account for 70% of the security issues in Microsoft products and similarly in Google Android. In this talk, we will introduce our research, which aims at providing architectural support for mitigating these security issues, on open-source RISC-V processor.
Tuo Li (M'15) is currently a postdoctoral researcher with the School of Computer Science and Engineering, University of New South Wales. His current research interests include timing side channel, memory safety, and fault tolerance for embedded systems and computer architecture. He received the B.E. degree in electronic science and technology from Hefei University of Technology, China, in 2008, and the Ph.D. degree in computer science and engineering from the University of New South Wales, Australia, in 2014. He has been a reviewer for journals such as IEEE TCAD, IEEE TVLSI, IEEE TDSC, IEEE Embedded System Letters, etc. He also has been a reviewer for security, hardware, and embedded system conferences such as CCS, DAC, ICCAD, DATE, ESWEEK, ASP-DAC, RTAS, RTSS. Tuo Li has been an active contributor in the RISC-V open-source community, where he maintains a RISC-V rocket chip port for Xilinx ZYNQ ultrascale+ FPGA board.
张洋 现任深圳大学新葡的京集团350vip8888副研究员 合作教授 沈琳琳
本科和博士（直攻博）就读于华中科技大学自动化学院，美国罗切斯特大学联合培养博士。长期研究忆阻神经网络、深度学习、模式识别，以独立第一作者身份发表中科院一区论文3篇，JCR一区论文6篇，IEEE trans系列论文7篇，在IEEE电路系统国际顶级会议报告2次，主持国家自然科学基金2项（青年基金、面上项目）、省部级科学基金3项与深圳市自然科学基金1项。授权专利4项，软著4部。担任Proceedings of the IEEE, IEEE Transactions on Cybernetics, IEEE Transactions on Fuzzy Systems, IEEE Transactions on Neural Networks and Learning Systems, IEEE Transactions on Industrial Electronics, IEEE Transactions on Circuits and Systems—I: Regular Papers, IEEE Transactions on Circuits and Systems II: Express Briefs, IEEE Transactions on Electron Devices, IEEE Electron Device Letters, Neurocomputing等期刊的独立审稿人。