
项目将首先回顾包含分类与回归的传统机器学习算法及初步神经网络,而后教授将会介绍用于优化神经网络的数学原理及代码技术。在确保学生具备扎实的理论及编程基础后,项目将进入到关于卷积神经网络原理、架构、优化及应用的核心阶段,学生将根据自身兴趣选择个性化研究课题进行深入研究,在项目结束时提交项目报告,进行成果展示。 In this course, you will be taken from basic topics of artificial neural networks to advanced topics such as convolutions. We will review important introductory concepts such as feedforward networks, gradient descent etc and then dive into convolutional neural networks. 个性化研究课题参考 Suggested Research Fields: 算法优化:图卷积神经网络 Graph Neural Networks 计算机视觉应用:DGD卷积神经网络行人重识别 Person re-identification on DGD convolutional neural networks 自然语言处理应用:基于自联想记忆与卷积神经网络的跨语言情感分类 Auto-associative convolutional neural network based multi-language sentiment classification 推荐系统应用:基于标签卷积神经网络的推荐算法 Personalised recommender system with tagged convolutional neural network
哈佛大学(Harvard University)始建于1636年,是一所享誉世界的私立研究型大学,也是常春藤盟校成员。哈佛大学在学术界享有崇高的地位,并且在世界范围内具有广泛的社会影响力。哈佛大学孕育了8位美国总统,158位诺贝尔奖获得者(世界第一)和18位菲尔兹奖得主(世界第一),在2019/2020年U.S.News世界大学排名中位列第一,2018年QS世界大学计算机科学以及电子工程专业排名位列第六。
经典机器学习算法回顾及神经网络初步 Introduction to Neural Networks, Review of Classification and Regression, and Simple Feed-Forward (FF) Network Neural, Network Architecture, Design Choices
梯度下降算法 Gradient Descent Algorithm
基于反向传播的自动微分算法 Automatic Differentiation using Backpropagation
神经网络优化技术 Neural Network Optimizers
神经网络正则化在防过拟合中的应用 Regularization for Neural Networks
卷积神经网络基本概念和体系结构 Convolutional Neural Networks: Basic Concepts, Padding, Pooling, and CNN Architecture.
感知野与通过池化层的反向传播 Receptive Fields, Backprop through max-pooling
显著图与神经网络最新技术展望 Saliency Maps State Of The Art network
项目回顾与成果展示 Program Review and Final Presentation
论文辅导与投递 Project Deliverables Tutoring


工程
黑碳的环境老化与环境氧化
同济大学 环境科学与工程学院 教授 博导 美国加州大学洛杉矶分校 访问学者 研究方向: 1. 环境污染与修复 2. 土壤环境学 3. 纳米环境学 主讲课程: 环境污染修复学,环境地学 已在专业核心学术刊物及国内外会议发表论文30余篇,申请国家专利6项,其中已授权4项。主持国家自然科学基金项目6项、浙江省自然科学基金1项,参与国家自然科学重点基金、国家重点研发计划项目课题、教育部科技创新工程重大培育项目、上海市科委社会发展项目4项。担任《农业资源与环境学报》编委、ES&T等二十余种SCI刊物审稿人,国家自然科学基金项目通讯评审专家、国家重点研发项目视频答辩评审专家。
工程
自动化与控制理论专题:机器人设计与应用研究【大学组】
Nader导师于1988年加入佐治亚理工学院,现任George W. Woodruff机械工程学院终身教授及机器人博士项目主任,Nader教授在加州大学伯克利分校取得硕士及博士学位,早期研究工作是在机器人和自动化领域。他在这一领域的主要贡献是开发了一类用于非线性机械系统的自适应和学习控制器,包括机器人操纵器,它使机器人像人类一样能够通过实践来学习重复的任务,且不需要精确的模型。后来工业生产证明,实现这种学习控制器可以在不显著增加成本或复杂性的情况下显著提高工业机器人的性能,并有潜力提高自动化制造系统的准确性、自主性和生产率。除了机器人技术,导师还曾开发了一种类似的学习控制器,用于调节复印机感光器的速度,这是施乐公司赞助的一个项目的一部分。 Dr. Nader's early research work was in the field of robotics and automation. His major contribution to this field was the development of a class of adaptive and learning controllers for nonlinear mechanical systems including robotic manipulators. This work, which evolved from his doctoral research, enables a robot to learn a repetitive task through practice, much like a human being, and without requiring a precise model. He later demonstrated that implementing this learning controller can significantly improve the performance of industrial robots without significantly increasing their cost or complexity, and has the potential to improve the accuracy, autonomy, and productivity of automated manufacturing systems. In addition to robotics, he developed a similar learning controller for speed regulation of copier photoreceptors as part of a project sponsored by the Xerox Corporation. Dr. Sadegh began at Tech in 1988 as an Assistant Professor.

自然科学
应用数学前沿研究【高中组】
Prof. Alberto is a Senior Faculty Scientist in the Mathematics Group at LBNL. He also holds a full-time position in the Mathematics Department at UC Berkeley. Before joining LBNL in 1976, Grünbaum taught at the Courant Institute in NYU, was a research scientist at the IBM Research Center in Yorktown Heights, NY and then joined the Applied Mathematics Department at Caltech before moving to Berkeley in 1974. He has served as director of the Center for Pure and Applied Mathematics and then Chairman of the UC Berkeley Mathematics Department from 1989--1992. He has been published extensively in the area of imaging, and his recent research interests include the study of Quantum Walks and its applications to different areas of science and technology. Alberto导师是加州大学伯克利分校应用数学终身正教授,在加州大学伯克利分校讲授线性代数等课程,曾任加州大学伯克利分校数学系主任、曾任英国物理研究所出版刊物Inverse Problems主编,曾在纽约大学柯朗数学研究所(Courant Institute;全球Top1应用数学研究中心)、IBM全球研究中心、劳伦斯伯克利国家实验室(Lawrence Berkeley Lab;美国最杰出的国家实验室之一)进行教学或研究工作。Alberto导师的研究聚焦应用数学与数学分析,多次应邀至世界各地知名学府发表主旨演讲。