性别：男
电话：18840867066
邮箱：[email protected]
学校：大连理工大学·软件工程系·数字媒体专业
-
大一上学期刚刚接触编程便产生浓厚兴趣，课余时间学习数学建模，并在寒假期间完成第一个小项目-数独，程序使用幻方变换生成数独初盘，使用深度优先搜索进行数字去除和终盘有效性检验，使用
c语言并于控制台下进行编程。 -
大一下学期接触优化问题和启发式算法，动手编写了遗传、粒子群优化等算法，并完成一篇基于列维飞行的粒子群算法优化论文的撰写和实验。
-
大二上学期参加微软创新杯完成一款关于跑步的社交软件(
c#&wp开发)——一起锻炼吧，之后写了一段时间的wp开发，有一些小作品。（其实这段时间不知道学什么，幸亏之后找到了正确的方向。） -
大二下学期开始接触机器学习并深入研究，刚开始通过学习Ng斯坦福公开课中经典算法，并用
matlab实现了机器学习实战一书上的大部分经典算法。之后开始补充线代、矩阵和概率统计方面数学基础（数学太重要!），并参加淘宝大数据比赛进行学习推荐系统方面的算法和实战，不仅对协同过滤、SVD、关联分析等推荐算法深入理解并通过实际数据在matlab上进行实验。 -
大三继续深入学习机器学习算法的理论，加强各个算法的推导过程和相互联系，对算法理论进行系统得整理并写博客。
- 进行一个基于wifi进行动作识别的项目，该项目通过csi信号收集动作信号，使用最近邻算法提取动作的特征模式，使用特征选择算法进行特征学习，并使用支持向量机和LDA进行分类器训练，最终使用交叉检验进行测试。
- 理论方面深入研究多视图学习这一课题，多视图学习分为协同训练、多核学习和子空间学习，协同训练方面主要是1998年提出的co-training算法，之后根据不同的潜在学习器的结合，学习多个视图的潜在知识。多核学习通过每个核代表每个视图，通过核的各种结合方式来进行视图的学习。子空间学习通过学习多视图共享的隐藏空间。
-
大学三年在学业方面最大的付出和收获都来自于机器学习，通过对相关理论的研究，基础算法的学习推导和实验，对机器学习打下基础，并且结合项目的真实数据和具体课题的深入研究，加强对算法的理解和处理现实情况的能力。之后会继续深入系统的理论学习，并希望在今后的实习，研究和工作阶段都能从事相关方向。
Matlab：科研实验
python：scikit-learn机器学习算法包
C/C++
JavaScript：unity 3D
C# & XAML：WP项目开发
机器学习：重要算法如SVM、EM的理论推导、证明及编码
数学储备：线性代数、概率统计、矩阵分析、凸优化问题
科研储备：英文文献阅读、理论推导、编码实验、综述或论文撰写（已具备做科研的耐心和部分能力）
-
数据结构算法可视化：将数据结构与算法中的线性表、树、图等经典算法用网页可视化表示。绘图层使用
javascript中的canvas编写绘图接口，算法层中进行具体的算法步骤编码并通过字符串的形式存储绘图命令，控制层接收绘图命令并调用绘图层中的接口，通过刷新画布进行图形的动态绘制，web前端使用html+css进行网页设计。我在项目中的任务是所有绘图层接口和控制层控制函数的编写，并完成算法层中线性表和树的部分算法编写。 -
基于WiFi信号的动作识别：基于Linux10操作系统，通过与路由器之间建立连接发送与接收CSI信号。在实验中被观测者做蹲下、跳跃、拾取等动作，通过捕获的CSI信号的变化曲线，使用
MATLAB编码，经过滤波、异常点检测、异常模式提取、特征计算、构建分类器与交叉检验等步骤，对不同的动作进行分类。我在项目中的任务是所用到的算法（如异常点检测、分类器算法等）的编码。 -
机器学习与多视图学习：使用
MATLAB和python两种语言实现机器学习中的knn、决策树、朴素贝叶斯、logistic回归、支持向量机、CART、kmeans聚类、谱聚类、关联分析、PCA、LDA、SVD等经典算法，并在博客中记录学习感悟，主要参考书籍是《统计学习方法》、《机器学习实战》、《机器学习》及线代、概率相关数学书籍等。多视图学习方面实验运用到很多机器学习算法，练习之后更容易实现，并完成几篇多视图学习的文献综述。
- 大连理工大学科技创新、社会工作、学习一等奖学金
- 住友化学二等奖学金
- 大连市ACM程序竞赛二等奖
- 大连市工科数学竞赛一等奖
- 东北三省数模竞赛三等奖
- 国际大学生数模竞赛三等奖
- 赴井冈山社会实践活动校优秀团队一等奖
- 学院心理健康协会宣传部副部长
- 班级宣传委员
- 有耐心毅力，并且能细致考虑问题。
- 有将一个问题挖到最深处完全理解的强迫症，主要反映在学习算法必须“嚼烂”。
- 希望从事机器学习和数据挖掘方面的实习，加强对机器学习算法实际应用的能力。
- 希望在研究生期间能进行机器学习方面的偏应用方面的科学研究。
- 加强数学和机器学习理论知识，达成
数学->科研->应用。
Gender: Male
Telephone: 18840867066
Email: [email protected]
University: Dalian Technology of University·Software Engineering Department·Digital Media Professional
The key skill of my university experience is Machine Learning, I touch it when sophmore. At first, I only remember the process of the algorithm because I cannot understand the derivation. Then, I study math in this field such as matrix theory, probability statistics and a little convex optimization theory. When junior I can use my mathematical knowledge to deduce the algorithms. For a machine learning algorithm, I understand the idea borning, model eastablishing, solving and proofing, characteristic analysising, coding and experimenting. But it’s not enough. Machine Learning is not only set of algorithms but also the framework to solve problem. Now I concnetrate on the essence of problem, the difference and contact between algorithms and in deep the complexity and generalization ability. To put into practice, I study recommendation system and participate in “Tamll big data competition”, use statistical learning to complete “Wi-Fi movement recognization project”. To put into scientific researching, I study Multi-view Learning in SSDUT Wisdom Laboratory. I want to practise in company when senior because I want to contact big data and machine learning in contact with reality. Then study on machine laerning related field during the postgraduate.
- Mathematical Modeling Contest third provincial prize
- Dalian Engineering Mathematics Contest first municipal prize
- Dalian ACM Competition second municipal prize
- Technology Innovation Model in Dalian University of Technology
- The first Learning Scholarship in Dalian University of Technology
- Sumitomo Chemical Scholarship
- Jinggangshan social practice first municipal prize
- Heuristic Algorithm : We study scheduling problem on parallel machines, where each job is unrelated with others and can be processed by any machine, with the goal of minimizing makespan. We propose an improved algorithm with Levy Flight (LPSO). Levy Flight can avoid the shortcoming of PSO, i.e., falling into prematurity, has been improved effectively, by randomly. The computational results show that the proposed LPSO algorithm is more competitive than the classical one.
- Algorithm visualization : We demonstrate the classical algorithms from data structure in form of visualization under Web. The framework contain 3 layers. In graph layer(bottom layer), we use canvas module to construct interface of drawing. In algorithm layer(top layer), design algorithms algorithm process and drawing strategy, then store the drawing commands as string. In control layer(middle layer), accept drawing commands from top layer and call the drawing interface in bottom layer. The animations rely on refreshing canvas.
- Wi-Fi movement recognization : We use CSI signal to recognize movements. The compute pings to router and receive the signal that transmit by router with 3 MIMOs. Received signal will fluctuate when human do different actions. We use these signal fluctuation extract features and classify different movements. The process contains signal filtering, outlier detection, action pattern extraction, feature extraction, feature selection, classifier construction and cross validation. The classifier is SVM with quadratic polynomial kernel and the accuracy rate of classification under 13 movements is 81.72%.
- Multi-view Learning : Multi-view Learning is a branch of machine learning. Because data can be gained from different methods and angles, the feature of problem can be extended to saveral sets. The multi-view means feature from different sources. First, we can use co-training method to combine the compatibility of multi views then get the common potenial infomation. Second, we can use multi kernel method. Each kernel represents to a view and combining kernels appropriately may improve learning performance. Third, we can use subspace method to obtain a latent subspace shared by multi views. I extensively read all the papers in this field and intensively read saveral key papers. I complete the experiments in these paper and do a compare between them. I complete 3 overviews of 3 sub filed, and now I am concentrated on a overview of Multi-view Learning.