的乘法分解。介绍两种求解该问题的方法：
-
-
-- nonlinear Gauss-Seidel method
-- nonlinear SOR(Successive Over-Relaxation)-like scheme:LMaFit
-
-
-其中SOR方法是GS方法的拓展，区别仅在于SOR方法中对于**X**的更新加了权重，并对权值进行更新。
-
-###去中心式算法:
-
-
当矩阵规模大到一定程度时，集中式算法在计算能力上要求过高，普通计算机也许无法计算。这时，我们需要在由许多普通计算机作为节点组成的网络中运算，这需要实现去中心式计算。去中心式计算式很容易实现的，将
,
,
分别切块放在每个节点上，将
作为公共信息在网络各个邻居节点间交换，优化问题形式不变，但需要加上
}=\mathbf{X}_{(j)} "\mathbf{ X}_{(i)}=\mathbf{X}_{(j)}")
的约束。而这样一个约束就引出了另一个子问题：一致平均(average consensus)问题。
-
-关于一致平均问题的介绍请看：
-
-- [《对于一致平均问题的理解》](http://painterlin.com/2014/08/22/Average-Consensus.html)
-- [《动态一致平均问题的4篇论文》](http://painterlin.com/2014/08/31/Papers-about-average-consensus.html)
-
-
-
-
-
-
diff --git "a/_posts/\347\247\221\347\240\224/2014-07-27-norms-of-vector-and-matrix.md" "b/_posts/\347\247\221\347\240\224/2014-07-27-norms-of-vector-and-matrix.md"
deleted file mode 100755
index 01eec41c7..000000000
--- "a/_posts/\347\247\221\347\240\224/2014-07-27-norms-of-vector-and-matrix.md"
+++ /dev/null
@@ -1,30 +0,0 @@
----
-layout: post
-title: 各类范数
-category: 科研
-tags: Math
-keywords: 向量范数，矩阵范数
-description:
----
-
-###向量范数
-
-
-
-###矩阵范数
-
-
-
-###矩阵乘积的迹
-
-
-###特殊范数
-- 矩阵W的L2-1范数：
-
-
-###TV范数
-||L(x)||_1。 其中L是差分算子，x是某种数字信号，在一维情况下，如下所示：
-
-||L(x)||_1 = |x2-x1| + |x3-x2| + |x4-x3| + ……
-
-加TV范数的目的是为了使求得的去噪信号仍然具有分段连续的性质。因为差值的1范数说明差值稀疏，从而说明求得的信号分段连续。
diff --git "a/_posts/\347\247\221\347\240\224/2014-08-22-Average-Consensus.md" "b/_posts/\347\247\221\347\240\224/2014-08-22-Average-Consensus.md"
deleted file mode 100755
index 477138fa0..000000000
--- "a/_posts/\347\247\221\347\240\224/2014-08-22-Average-Consensus.md"
+++ /dev/null
@@ -1,55 +0,0 @@
----
-layout: post
-title: 对于一致平均问题的理解
-category: 科研
-tags: Optimization
-keywords: 矩阵补全，一致平均
-description:
----
-
-
-最近几个月在研究分布式低秩矩阵补全的问题，参考文章[《低秩矩阵补全》][1]。
-
-低秩矩阵补全问题的各类算法中，很关键的一个子问题叫做一致平均问题(Average Consensus)，而上一篇文章并没有对这个问题进行说明。那么，什么叫做“一致平均”呢？
-
-
-###问题阐释：
-
考虑一个有
个节点的网络，每个节点上存储一个关于自己的数值信息，叫做初值。一致平均问题就是使所有节点在算法停止的时候收敛到
个初值的平均。
-
-
-
在分布式算法中，通常会存在这样一个变量
，它作为公有信息在网络中传递，每个节点储存自己的
} "\mathbf{X}_{(i)}")
。在总算法的每一次迭代中，每个节点
接收自己邻居节点传递过来的公有信息，然后与自己的私有信息共同计算出新的公有信息，并传给自己的邻居。我们需要保证每个节点上的公有信息
，使分布的算法以某种方式交流合作，以便获得最优解。这就是分布式算法中的一致平均问题。
-
-###问题分类：
-根据节点上的数值信息是否随时间变化，又把一致平均问题分为：
-
-- 静态一致平均 (Static average consensus)
-- 动态一致平均 (Dynamic average consensus)
-
-
-顾名思义，静态一致平均指节点上的初值不会发生变化，只需要保证最后每个节点都收敛到
个初值的平均值即可；而动态一致平均则是，节点上的数值不断发生变化，即
在
时刻的值并不一定与0时刻的值（初值）相同，我们使用不断变化的公有信息（因为公有信息不断在被更新），仍需要保证最后每个节点都收敛到
个初值的平均。相比于静态一致平均，动态一致平均问题更为棘手。
-
-
-###求解方法分类：
-在文章《低秩矩阵补全》的最后，提供一种求解方法的分类概念，将方法分为：
-
-- 精确一致平均 (Exact average consensus)
-- 不精确一致平均 (Inexact average consensus)
-
-这又是什么意思呢？
-
-
-在解决分布式低秩矩阵补全问题的时候，我们将算法分解两个子问题不断求解，一是交替极小化(Alternating minimization)得到每个节点的新的
,
；二是在网络中对
个节点的
求一致平均。对于第二个子问题，在每一次算法总的迭代中，都去求解精确的一致平均显然能够解决问题，但是因为一致平均也需要一定次数的迭代才能被解除，如果在总算法的每一次迭代中都去求精确的一致平均，则相当耗费计算资源，增加了算法的时间复杂度。
-
-
-很自然地，我们会想到，既然一致平均只是矩阵补全的一个子问题，我们是不是可以通过某种松弛，来降低算法的时间复杂度并节省计算资源，同时仍旧保证总算法的收敛呢。这样，就提出了不精确的一致平均。
-
-
-不精确一致平均，就是在每一次求解一致平均子问题时，只迭代一次或者若干次，而不是迭代所有
次（
可以很大），使每个节点`近似`的达到它们初值的均值。
-
-
-###精确一致平均与不精确一致平均优缺点比较:
-
-- 精确(Exact)求解：理论上容易证明，但计算代价通常比inexact方法高
-- 不精确(Inexact)求解：理论分析上不好证明，除此之外具有exact不具有的所有优点，比如算法时间复杂度低，节省计算资源等等。
-
- [1]: http://painterlin.com/2014/07/25/matrix-completion.html
diff --git "a/_posts/\347\247\221\347\240\224/2014-08-31-Papers-about-average-consensus.md" "b/_posts/\347\247\221\347\240\224/2014-08-31-Papers-about-average-consensus.md"
deleted file mode 100755
index 8e6015896..000000000
--- "a/_posts/\347\247\221\347\240\224/2014-08-31-Papers-about-average-consensus.md"
+++ /dev/null
@@ -1,21 +0,0 @@
----
-layout: post
-title: 动态一致平均问题算法-EXTRA和DAC
-category: 科研
-tags: Optimization
-keywords: 矩阵补全，一致平均，论文
-description:
----
-
-### EXTRA
-
-
}^1=\sum_{j&space;=&space;1}^Nw_{ij}x_{(j)}^0-\alpha&space;\nabla&space;f_i(x_{(i)}^0)\\&space;\\&space;\\&space;x_{(i)}^{k+2}=x_{(i)}^{k+1}+\sum_{j=1}^Nw_{ij}x_{(j)}^{k+1}-\sum_{j=1}^N\tilde{w}_{ij}x_{(j)}^k-\alpha[\nabla&space;f_i(x_{(i)}^{k+1})-\nabla&space;f_i(x_{(i)}^{k})] "x_{(i)}^1=\sum_{j = 1}^Nw_{ij}x_{(j)}^0-\alpha \nabla f_i(x_{(i)}^0)\\ \\ \\ x_{(i)}^{k+2}=x_{(i)}^{k+1}+\sum_{j=1}^Nw_{ij}x_{(j)}^{k+1}-\sum_{j=1}^N\tilde{w}_{ij}x_{(j)}^k-\alpha[\nabla f_i(x_{(i)}^{k+1})-\nabla f_i(x_{(i)}^{k})]")
-
-### DAC
-
-
=x_i(t)+\sum_{j&space;\neq&space;i}a_{ij}(t)(x_j(t)-x_i(t))+r_i(t)-r_i(t-1) "x_i(t+1)=x_i(t)+\sum_{j \neq i}a_{ij}(t)(x_i(t)-x_i(t))+r_i(t)-r_i(t-1)")
-
-### 优缺点比较
-其中，DAC最大的缺点在于第一次迭代时对于r(-1)时刻的依赖，在实际仿真中，如果需要对动态输入求一致平均，往往并不能获取输入在-1时刻的值。导致在矩阵补全问题中，DAC做不精确的动态一致平均的子问题效果并不好。
-
-而EXTRA却有很好的效果。
\ No newline at end of file
diff --git "a/_posts/\347\247\221\347\240\224/2014-09-11-Matlab-tips.md" "b/_posts/\347\247\221\347\240\224/2014-09-11-Matlab-tips.md"
deleted file mode 100755
index 77eab3e95..000000000
--- "a/_posts/\347\247\221\347\240\224/2014-09-11-Matlab-tips.md"
+++ /dev/null
@@ -1,97 +0,0 @@
----
-layout: post
-title: Matlab科研小贴士
-category: 科研
-tags: Matlab
-keywords: 矩阵补全，Matlab
-description:
----
-
-使用MATLAB运行算法程序时，可能遇到各种各样的报错。比如，为了保护隐私数据，我在分布式矩阵补全算法中加入随机矩阵之后，某项变量在运行几百步之后会出现NaN报错。我只根据算法顺序去分析问题出现的可能原因，并修改程序。感觉并没有很好地利用MATLAB的强大功能去锁定症结所在。幸运的是，[施伟](http://home.ustc.edu.cn/~shiwei00/index.html)大师兄当时和我在一起，他非常热心地帮我分析问题，教我以后遇到类似状况应该怎么去分析与思考。和大师兄讨论了半小时，感觉自己收获不少。
-
-这篇文章会陆续记录下自己使用MATLAB的体会，以及解决问题的一些技巧。
-
-### Clear
-运行一段代码前通常需要将工作空间里的已有数据清除掉。只需要在编辑有实际意义的代码之前写下如下代码：
-
-```matlab
-clc; clear; close all;
-```
-
-### Random Seed
-为了保证程序在相同环境下运行以便测试某一个或几个改变对于算法的影响，在使用各种random命令时，需要设定固定种子。这样就不会因为每次随机产生的序列不同而影响程序运行结果。设置随机种子的代码如下：
-
-```matlab
-%% random seed
-%seed=round(5000*rand); % use this line if you set a random seed
-seed=3302; % use this line if you set a fixed seed. 3302 can be replaced by other numbers.
-fprintf('Seed = %d\n',seed); % print the current seed
-if exist('RandStream','file')
- RandStream.setGlobalStream(RandStream('mt19937ar','seed',seed));
-else
- rand('state',seed);
- randn('state',seed^2);
-end
-```
-
-### NaN
-NaN是Not a Number的缩写。当某变量显示NaN时，表示该变量是不明确的数值结果。比如0/0、inf/inf等运算会出现NaN报错。遇到这种情况，首先判断NaN出现在哪一步：
-
-
-```matlab
-if isnan(norres) %括号里是变量名。判断norres是否为NaN，若是，则在该步暂停程序。
- keyboard;
-end
-```
-再在命令窗里单独查看与该变量有关的其他变量，从而排除正常变量，获知究竟是哪个或哪几个变量出了问题，变为无穷大或无穷小。再检查与这些变量有关的算法。
-
-
-### SaveAs
-若需要比较各参数对算法性能的影响，通常是在程序中修改参数运行，得到算法收敛精度与迭代次数的曲线图。再根据曲线图反向思考修改哪些参数有效。这个过程需要保存产生的大量图片。可以使用`hold on`命令将所有虚线画在同一张图上，也可以使用`saveas`将所有图片自动保存。
-
-
-```matlab
-%% plot
-figure(1)
-semilogy(1:iter,y_axis(1:iter),'b-'); %b：蓝色。－：线段形状
-set(gca,'fontsize',12);
-grid on;
-xlabel('\fontsize{12}\it Iteration'); ylabel('\fontsize{12}\it Normalized residual');
-legend('\fontsize{12}\it text'); %text：这条蓝色代表什么
-hold on
-saveas(gcf,'filename','fig') %filename：将图片保存为这个名字。fig：保存为fig格式
-```
-
-保存变量数据的命令：
-
-```
-save('filename')
-save('filename','variables')
-```
-
-注意，在使用`hold on`命令时，应该保留上次程序运行后产生的各种数据。即不能在程序中写类似与`clear all`之类的清除语句，否则上次曲线图也将被删除。
-
-### 矩阵规范化
-已知满秩矩阵A，进行下面操作使其所有奇异值均为1。
-
-```matlab
-[u s v]=svd(A);
-A=u*v';
-```
-
-### 安装CVX
-
-1. 将cvx压缩包解压
-2. 将cvx文件夹拷贝至如D:\MATLAB Programs\Compressed Sensing目录下
-3. 在Current Folder窗口中打开cvx文件夹
-4. 在Command Window中输入cvx_setup
-5. 在MATLAB的File菜单下的set path把此路径加上。
-6. 把路径加进去后在file→Preferences→General的Toolbox Path Caching里点击update Toolbox Path Cache更新一下
-7. 完成
-
-### %%分段运行程序
-
-1. 选中%%分段
-2. 右键选择`evaluate current section`
-
-
diff --git "a/_posts/\347\247\221\347\240\224/2014-10-01-Alibaba-Big-Data.md" "b/_posts/\347\247\221\347\240\224/2014-10-01-Alibaba-Big-Data.md"
deleted file mode 100755
index b81494539..000000000
--- "a/_posts/\347\247\221\347\240\224/2014-10-01-Alibaba-Big-Data.md"
+++ /dev/null
@@ -1,99 +0,0 @@
----
-layout: post
-title: 阿里巴巴大数据竞赛回顾与总结
-category: 科研
-tags: MachineLearning
-keywords: 大数据，机器学习，推荐系统
-description:
----
-
-8个月前，苏宇师兄问我对[阿里巴巴大数据竞赛](http://102.alibaba.com/competition/addDiscovery/index.htm)感兴趣吗。正好我选修了[陈恩红](http://staff.ustc.edu.cn/~cheneh/)老师的《机器学习与数据挖掘》，很好奇这门课在实际中的应用；身为淘宝用户，也好奇我是如何被推荐的。于是和师兄一起参加比赛，加上计算机学院的刘惠民同学，我们组成了名叫`Rosemary`三人团队。这次大赛总共有7276支队伍参赛，我们止步于第二赛季。第一赛季排名56；第二赛季排名68。
-
-###开放数据
-
-
-
- | 字段 |
- 字段说明 |
- 提取说明 |
-
-
- | user_id |
- 用户标记 |
- 抽样、字段加密 |
-
-
- | Time |
- 行为时间 |
- 精度到天级别、隐藏年份 |
-
-
- | action_type |
- 用户对品牌的行为类型 |
- 包括点击（0）、购买（1）、加入购物车（2）、收藏（3）四种行为 |
-
-
- | brand_id |
- 品牌数字ID |
- 抽样、字段加密 |
-
-
-###比赛任务
-根据用户4个月在天猫的行为日志，建立用户的品牌偏好，并预测他们在将来一个月内对品牌下商品的购买行为。
-
-###评估指标
-大赛最终的比赛成绩排名以F1得分为准。
-
-准确率：
-
-召回率：
-
-F1-Score：
-
-其中，
-
-- N 为参赛队预测的用户数；M 为实际产生成交的用户数量
-- pBrandsi为对用户i 预测他(她)会购买的品牌列表个数；bBrandsi为用户i 真实购买的品牌个数
-- hitBrandsi对用户i预测的品牌列表与用户i真实购买的品牌交集的个数
-
-###解读`准确率`、`召回率`和`F1-Score`
-准确率就是正确预测数目比上预测总数目。召回率就是正确预测数目比上真实总数目。F1-Score是准确率和召回率的调和平均。理论上，准确率与召回率并没有必然的联系；但在实际中，这二者往往此消彼长、相互制约。有研究表明，在不牺牲准确率的情况下，获得一个高召回率是很难的。在赛题环境下举个栗子：
-
-
-
- | 我们预测出 |
- 会有3个人买东西，A买品牌a，B买品牌b，C买P品牌c |
-
-
- | 真实情况是 |
- 有4个人买了东西，这4个人分别是A，B，C，D，其中A买a和b，B买b，C买b和c，D买a。 |
-
-
-
-按照我们的预测，准确率达到了100%，而召回率仅为50%。由此可见，想取得一个较好的F1-Score成绩，需要保证预测的数据尽量hit之外，还要保证预测的数据能够覆盖尽量多的真实数据。具体到比赛中，就需要很好地确定每个被预测的用户可能会买多少个品牌。
-
-通过对这三个评判标准的分析，我们初步确定了第一赛季所用的策略。
-
-###第一赛季
-大致的方案是：
-
-1. 加权对用户4种行为次数求和，得到简单的用户ID（行）与品牌ID（列）的`兴趣度`（值）矩阵。
-2. 统计前4个月平均每个月有多少用户购买超过2个品牌的东西，记为B。
-3. 根据用户前4个月的购买行为，按照`购买力`划分用户购买等级。直白解释就是你以前在天猫上平均每个月买了多少次，我就假设你下个月还会买这么多次。
-4. 第一次切分兴趣度矩阵，保留前B行矩阵。
-3. 对每个用户对不同品牌的兴趣度由高到低排序，再根据每个用户的购买力确定对该用户保留前n列的兴趣数据，并预测他会买这n个数据对应的品牌。
-
-当然，在最初的模型建立起来之后，我们还陆续加了一些规则，比如在`时间节截止点`(最后N天，N可以根据自己的理解取值)的单独分析。另外，调参数真就是一项体力活。
-
-
-
-###第二赛季
-
-（好困，需要去睡觉Zzz···）细节未完待续
-
-
-
-
-###总结
-
-
diff --git "a/_posts/\347\247\221\347\240\224/2014-11-09-What-is-NP-Hard.md" "b/_posts/\347\247\221\347\240\224/2014-11-09-What-is-NP-Hard.md"
deleted file mode 100755
index 8ae62a5d1..000000000
--- "a/_posts/\347\247\221\347\240\224/2014-11-09-What-is-NP-Hard.md"
+++ /dev/null
@@ -1,51 +0,0 @@
----
-layout: post
-title: 什么是 P, NP, NP-complete, NP-hard
-category: 科研
-tags: Math
-keywords: 概念，NPhard
-description:
----
-
-###相关概念
-NP-hard（non-deterministic polynomial-time hard）
-
-1. P：能在多项式时间内解决
-2. NP：不能在多项式时间内解决或不确定能不能在多项式时间内解决，但一旦你找到一个解，只需要多项式时间去验证这个解是正确的
-3. NP-hard：如果一个问题是NP-hard，意味着可以将任意NP问题化约到这个问题。如果可以解这个问题，那么可以轻松地解任意NP问题。
-4. NPC：NP完全问题，所有NP问题在多项式时间内都能化约（Reducibility）到某一NP问题，这一NP问题就是NPC问题，即解决了此NPC问题，所有NP问题也都解决了。
-
-###资料原文
-These refer to how long it takes a program to run. Problems in class P can be solved with algorithms that run in **polynomial time**.
-
-Say you have an algorithm that finds the smallest integer in an array. One way to do this is by iterating over all the integers of the array and keeping track of the smallest number you've seen up to that point. Every time you look at an element, you compare it to the current minimum, and if it's smaller, you update the minimum.
-
-How long does this take? Let's say there are n elements in the array. For every element the algorithm has to perform a constant number of operations. Therefore we can say that the algorithm runs in O(n) time, or that the runtime is a linear function of how many elements are in the array. So this algorithm runs in **linear time**.
-
-You can also have algorithms that run in **quadratic time** (O(n^2)), **exponential time** (O(2^n)), or even **logarithmic time** (O(log n)). Binary search (on a balanced tree) runs in logarithmic time because the height of the binary search tree is a logarithmic function of the number of elements in the tree.
-
-If the running time is some polynomial function of the size of the input, for instance if the algorithm runs in linear time or quadratic time or cubic time, then we say the algorithm runs in **polynomial time** and the problem it solves is in class **P**.
-
-###**NP**
-Now there are a lot of programs that don't (necessarily) run in polynomial time on a regular computer, but do run in polynomial time on a nondeterministic Turing machine. These programs solve problems in **NP**, which stands for **nondeterministic polynomial time**. A nondeterministic Turing machine can do everything a regular computer can and more. This means all problems in P are also in NP.
-
-An equivalent way to define NP is by pointing to the problems that can be verified in polynomial time. This means there is not necessarily a polynomial-time way to find a solution, but once you have a solution it only takes polynomial time to verify that it is correct.
-
-Some people think P = NP, which means any problem that can be verified in polynomial time can also be solved in polynomial time and vice versa. If they could prove this, it would revolutionize computer science because people would be able to construct faster algorithms for a lot of important problems.
-
-###**NP-hard**
-What does NP-hard mean? A lot of times you can solve a problem by reducing it to a different problem. I can reduce Problem B to Problem A if, given a solution to Problem A, I can easily construct a solution to Problem B. (In this case, "easily" means "in polynomial time.")
-
-If a problem is **NP-hard**, this means I can reduce any problem in NP to that problem. This means if I can solve that problem, I can easily solve any problem in NP. If we could solve an NP-hard problem in polynomial time, this would prove P = NP.
-
-###**NP-complete**
-A problem is **NP-complete** if the problem is both
-
-- NP-hard, and
-- in NP.
-
-###参考资料
-【1】[What are P, NP, NP-complete, and NP-hard?](http://www.quora.com/What-are-P-NP-NP-complete-and-NP-hard)
-
-
-
diff --git "a/_posts/\347\247\221\347\240\224/2014-11-10-Basic-Algorithms-of-Convex-Opt.md" "b/_posts/\347\247\221\347\240\224/2014-11-10-Basic-Algorithms-of-Convex-Opt.md"
deleted file mode 100755
index b4ba6b9ea..000000000
--- "a/_posts/\347\247\221\347\240\224/2014-11-10-Basic-Algorithms-of-Convex-Opt.md"
+++ /dev/null
@@ -1,27 +0,0 @@
----
-layout: post
-title: 凸优化的一些基础算法
-category: 科研
-tags: Optimization
-keywords: 凸优化，基本算法
-description:
----
-
-本文假设读者对凸优化有基本了解，主要归纳一些基础算法，以便查阅。
-
-
&space;\triangleq&space;g(x)+h(x) "\min f(x) \triangleq g(x)+h(x)")
-
-其中，f，g，h都是凸函数，g是光滑项，h是非光滑项。
-
-###Gradient Descent
-
 "x^+=x-\alpha\nabla f(x)")
-###Proximal Gradient
-
)=&space;\arg\min_u&space;h(u)+\frac{1}{2\alpha}||u-x+\alpha&space;\nabla&space;g(x)|| "x^+=prox_{\alpha h}(x-\alpha \nabla g(x))= \arg\min_u h(u)+\frac{1}{2\alpha}||u-x+\alpha \nabla g(x)||")
-###Conjugate Gradient
-是介于最速下降法和牛顿法之间的一个方法，它仅需要利用一阶导数信息，但克服了最速下降法收敛慢的缺点，又避免了牛顿法需要存储和计算Hession并求逆的缺点。它是解决大型线性方程组最有用的方法之一，也是解决大型非线性最优化最有效的算法之一。
-
-###Newton
-见[牛顿法与拟牛顿法（DFP BFGS LBFGS VLBFGS）](http://painterlin.com/2015/03/23/Newton-QuasiNewton-Method.html)
-
-###Quasi Newton
-见[牛顿法与拟牛顿法（DFP BFGS LBFGS VLBFGS）](http://painterlin.com/2015/03/23/Newton-QuasiNewton-Method.html)
diff --git "a/_posts/\347\247\221\347\240\224/2014-11-26-Presentation-at-Chinese-Academy-of-Science.md" "b/_posts/\347\247\221\347\240\224/2014-11-26-Presentation-at-Chinese-Academy-of-Science.md"
deleted file mode 100755
index 99b8fea14..000000000
--- "a/_posts/\347\247\221\347\240\224/2014-11-26-Presentation-at-Chinese-Academy-of-Science.md"
+++ /dev/null
@@ -1,207 +0,0 @@
----
-layout: post
-title: 我的报告 Decentralized Privacy-Preserving Low-Rank Matrix Completion
-category: 科研
-tags: Math Optimization
-keywords: 报告总结
-description:
----
-
-
-
-
-
-
-###我的报告
-
-####Section 0: Introduce Myself
-Page 1
-
-- Good afternoon everyone! My name is Anya Lin. I'm a second-year master candidate from University of Science and Technology of China. It's my great pleasure to introduce to you the Decentralized Privacy-Preserving Low-Rank Matrix Completion. It's a joint work with my supervisor, Prof. Ling from USTC.
-- Before I start, I want to express my thanks to Prof. Ling for his patient instructions and help over the last a few months.
-
-Page 2
-
-- Here's the outline of my presentation.
-- First is the introduction.
-- And then the centralized matrix completion problem.
-- We develop a decentralized algorithm in section 3, and our algorithm is derived from a centralized algorithm as I will talk about in section 2.
-- Next, I will introduce the main result of the topology-dependent privacy preservation.
-- At last, it's the conclusion.
-
-####Section 1: Introduction
-Page 3
-
-- OK， let's go into the introduction.
-
-Page 4
-
-- I'd like to begin with the concept of matrix completion. So what is matrix completion? As we can see in this picture, we have an incomplete matrix, whose entries are known only for a subset of the whole matrix. And the rank of the matrix is very small compared with the size of the matrix.
-- The goal of the matrix completion is to recover all these unknown entries of of the matrix, as the right-side picture shows. Here, Z is the recovery of W.
-
-Page 5
-
-- There's many applications of such a problem. Like image processing, recommendation system and so on. Here are 2 examples. The first one is a problem of image processing. The left picture has a lot of noises, or say, only a part of the original picture is known. By using the fact that the original picture is usually low-rank, we can matrix completion to denoise the picture and get a clear version of high quality as the right picture shows.
-- The second example is more close to our lives. It's related to a recommendation system. As you can see, it's a webpage of Douban Movie. A user sees a movie, such as Interstella, and then scores it on the website. Here we can imagine a huge matrix with rows representing users and columns the movies. This matrix is incomplete and it's low-rank. Once this matrix is completed, the website can recommend new movies to users.
-
-Page 6
-
-- Now, here comes a privacy concern. First what is privacy? Privacy is the values one considers private. In the example we mentioned just now, the users' scores of the movies are privacy, because one may not want others to know what movies he has seen or likes. Also, the entries of the matrix could be medical records of patients, or selling data of merchants. These data are considered as privacy.
-- Obviously, no one wants the leakage of his privacy.
-- However, in reality there may exist a malicious agent, a bad guy. For some reasons you have to give your private data to it, but you don't know wether you can trust it or not.
-- In this situation, we need privacy-preservation. Privacy-preservation is the ability to prevent a malicious agent from obtaining or reconstructing the private data.
-
-####Section 2: Centralized Matrix Completion Problem
-Page 7
-
-- Now let's go into the centralized matrix completion problem.
-
-Page 8
-
-- When we are faced with a low-rank matrix completion problem, the intuitive thought would be to minimize the rank of the matrix, but this is a nonconvex problem.
-- Therefore, we insteadly minimize the nuclear-norm the the matrix, since nuclear-norm is the approximation of the rank and it's convex.
-- Another approach is if the rank of matrix is known to be r as a prior theoreticallyor empirically, we can get the matrix factorization formulation. This approach is advantageous over the nuclear-norm approach since the latter one needs sigular value decomposition, which is computationally expensive and even intractable in decentralized computing.
-- A centralized algorithm called LMaFit to solve this is as the following steps shows.
-- We have to keep in mind that our algorithm is derived from LMaFit.
-
-####Section 3: Decentralized Matrix Completion Problem
-Page 9
-
-- After the centralized problem, let's go into the decentralized one.
-
-Page 10
-
-- In decentralized computing, we have a network composed of L agents.
-- There is an undirected edge between two agents if they can communicate with each other through one hop.
-- The goal of all the agents in such a network is to collaboratively complete a low-rank matrix in a decentralized fashion.
-
-Page 11
-
-- To be specific, we segment the whole data matrix W into groups of columns. And do the same to Y and Z. Each agent i in the network holds the corresponding Zi, Yi and entries of Wi.
-- However, X cannot be segmented and distributed to agents because the update of X contains the summation of ZiYi' over all agents. So we let each agent i holds a local copy X(i) of X.
-- After doing this, we get a naive decentralized implementation of LMaFit. At iteration k, each agent i does the following steps respectively.
-- Notice that the update of X requires information aggregation of all agents. So here is the challenge: informationaggregationofallagentsisimpossible in real decentralized network unless every agent is connected to all the other agents.
-- How to deal with this challenge?
-
-Page 12
-
-- The answer is dynamic average consensus. Recall that each agent i holds a local copy X(i) of X. If we can make sure that X(i) equals to X for all i, the challenge is solved.
-- To do this, we choose c to be 1/L and the update of X becomes the average consensus problem, as we can see in equation (8), X(i) is the average of all the ZiYi'.
-- At iteration k,we formulate the average consensus problem as equation (9). The constraint means that instead of letting all the X(i) to be identical we choose to let each X(i) equals to it neighboring X(j).
-- A key observation is that exact average consensus at every iteration is not necessary. We use EXTRA to do inexact dynamic average consensus, which saves the computational cost.
-
-Page 13
-
-- Our decentralized algorithm called D-LMaFit is developed as below.
-- Step 1 is the initialization.
-- Step 2, use EXTRA to do the inexact average consensus problem.
-- Step 3, update Y and X respectively.
-
-Page 14
-
-- The performance of D-LMaFit is shown in these two pictures. (Explain what these pictures indicate to the audience)
-
-####Section 4: Topology-Dependent Privacy Preservation
-Page 15
-
-- Now let's go into the section of the topology-dependent privacy preservation.
-
-Page 16
-
-- First compare decentralized algorithm with centralized one.
-- Centralized algorithm needs a fusion center to collect global data. What if the fusion center is a malicious agent?
-- Oops TT, you'll lose all your privacy.
-- How about the decentralzied algorithm? One important advantage of decentralized algortihm over a centralized is there's no global data collection, each agent observes part of the raw data and communicates with its neighboring agent(s). It seems safer.
-- But things aren't so lucky in reality. Because the communication of X(i) among the network may lead to information leakage.
-
-Page 17
-
-- How does this happen? Suppose in a network as this picture shows, we have a malicious agent M, and M attempts to recover the local data matrices of some other agents through information exchange.
-- M is interested in recovering the local data Wi, or equivalently Yi or Zi of a set of agents i∈I.
-- When the iteration k is large enough, X(i) will be identical. So if a malicious agent M somehow knows other agents' Yi, it can recover the data Zi of agent i by doing X(M)Yi.
-- So our concern is, is there any possibility that the malicious agent M can somehow obtain Yi of agent i, and thus get Zi, which is private.
-
-Page 18
-
-- Before going to details, consider two simple topologies.
-- (Explain the two topologies)
-- Under what conditions can not a malicious agent M reconstruct the sensitive information of P and Q ?
-
-Page 19
-
-- Recall the update of X.
-- If you could just take a look at the equation, you can find that if the topology is as in the left picture, M can reconstruct ZiYi' and it may be able to solve Yi so that gets the privacy of P and Q.
-- But if the topology is as shown in the right picture. M cannot get the private data of P and Q. Why?
-- (Explain with the equation)
-- M can solve a series of linear inverse problems and calculate the values of ZiYi', as what we have said just now.
-
-Page 20
-
-- Now we get a naive conclusion.
-
-Page 21
-
-- So the privacy-preserving problem boils down to the linear inverse problem. First we define some variables as this.
-- And further we define A and B.
-- Using these definition,the update of X can be represented by (14).
-
-Page 22
-
-- Rewrite this as a linear time-invariant systems we get (15). In this system, QI selects those row blocks in Q belonging to the agents in I, and BI selects the corresponding columns in B. QIC and BIC selects the other corresponding row blocks and columns which do not belong to the agents in I.
-- Our analysis uses the concept of z-transfer matrix of (15). The concept is from modern control theory.
-- Obviously, rank(T)=rank(TI TIC), since the latter matrix is just a column rearrangement of the former one.
-- Now we are ready to develop our theorem.
-
-Page 23
-
-- Check the proof of the suffienciency of our theorem, it's rather straightforward.
-- If this condition is satisfired, then M has full knowledge of all the X(i).
-- So M can solve a series of linear inverse problems.
-
-Page 24
-
-- The proof of necessity is a little bit complicated. Here's the only the simplified version of the proof.
-- First we show that to determine a unique sequence of Q from V , we must have (18).
-- Suppose (18) doesn't hold, then there exists at least one column of TI that is linearly dependent on the other columns of T. Then there exists a Q with that column nonzero, and satisfies TQ=0. This corresponds to a nonzero input in I, but the output V is zero for all time. Thus this nonzero input cannot be recovered. This contradicts with the hypothesis. So (18) must hold.
-- (Explain these items)
-
-Page 25
-
-- (Explain these items)
-
-####Section 5: Conclusion
-Page 26
-
-- I'd like to quickly go over the main point of today's topic.
-
-Page 27
-
-- First, we propose a decentralized privacy-preserving algorithm, D-LMaFit, to solve the matrix completion problem.
-- We solve dynamic average consensus subproblem inexactly.
-- We prove the topology-dependent privacy-preserving theorem. It provides a guideline of designing a privacy-preserving network.
-
-Page 28
-
-- Still we've got work to do in the future. (Read items)
-
-Page 29
-
-- I guess that's it. Thank you all very much for listening. Now if you have any question, please feel free to ask me.
-
-
-###故事
-这学期我在中科院数学与系统科学研究院(AMSS)访问。第一次参与这边的讨论班时，我就被惊到了：学生做报告也全程英文，不愧是[袁亚湘](http://lsec.cc.ac.cn/~yyx/chinese/indexc.htm)老师的学生。于是，11月25日，我也在这儿完成了自己第一次的英文学术报告。
-
-1. 报告前3天，我问盛镇醴师兄他们报告前会不会排练，师兄说：“肯定要啊！上次去葡萄牙开会，[马士谦](http://www.se.cuhk.edu.hk/people/sqma.html)师兄已经讲得那么好了，都还又自己私下练习了5、6遍。师兄真的可以做到每句话精确到几秒钟！”太荔枝了有木有TT。
-2. 于是我也练习了。果然只有努力了内心才会踏实。在当天的报告中，我不仅不紧张，还在瞅到台下一堆人的专注神情时，心里突然弹幕全开：“哇，这感觉好爽。”
-
-在记录报告之前，插播一段回忆：
-大三暑假，我参加[中国大学生物联网创新创业大赛](http://china.ican-contest.org/index.html)，正式比赛前一天系里组织答辩练习，我们组讲得一塌糊涂。那一晚，我和向国菲师兄在实验室通宵改幻灯片，准备发言稿，然后一句一句地练习。中途师兄压力太大又累得不行溜出去躲着抽了根烟，回来被我发现了教育了一顿，嗅觉就是这么灵敏没办法。直到凌晨4点，终于觉得还算满意了，两人躺椅子上睡了会儿，当然我被蚊子咬安逸了。早晨7点，寝室开门，两人各自回去洗澡调整状态。9点，开始比赛。不知道为什么突然想起这个，太，美好了。尽管当时觉得真苦逼。
-
-
-
-###资料
-【1】[有哪些高级的英语表达技巧，让人一听就很地道？](http://www.zhihu.com/question/24544386/answer/30237316)
-
-
-
diff --git "a/_posts/\347\247\221\347\240\224/2015-02-04-Group-Meeting.md" "b/_posts/\347\247\221\347\240\224/2015-02-04-Group-Meeting.md"
deleted file mode 100755
index d9bec4657..000000000
--- "a/_posts/\347\247\221\347\240\224/2015-02-04-Group-Meeting.md"
+++ /dev/null
@@ -1,92 +0,0 @@
----
-layout: post
-title: 大数据机器学习初探---南大李武军
-category: 科研
-tags: Math Optimization MachineLearning
-keywords: 组会
-description:
----
-
-每周的组会大概会持续2小时。如果是主讲，就需要花更多的时间去准备报告内容。之前，组会开完我就不管了，缺乏总结思考。而这样子的话实质上意义就不大了，没有内化为自己的知识，也没有什么critical thinking。从现在开始，记录每一次组会的思考。
-
-常言道：亡羊补牢，为时未晚。T.T
-
-###Outline
-- Learning to Hash
-- Distributed Learning
-- Stochastic Learning
-
-有一个形象的比喻是这样说的，大数据是金矿，云计算是采矿技术，大数据机器学习是冶金技术。
-
-大数据机器学习面临的挑战，一是存储，而是计算速度，三是网络。
-
-哈希学习，在内存、硬盘、cpu、通信上有优势；
-分布式学习在内存、硬盘、cpu上有优势，但会增加通信成本；随机学习在内存、硬盘、cpu方面有优势。
-
-
-###Learning to Hash
-主讲人：大师兄
-
-最近邻搜索在大数据背景下，会出现维数灾难，存储成本也高，查询速度也慢。解决方法之一是保相似性哈希，可以降低维数并减少存储成本。通常用海明距离（hamming distance）来表征哈希值之间的差异。哈希方案也具有较快的查询速度，通常具有常数或者次线性的搜索时间复杂度；即使是穷举搜索也可以被接受，因为海明距离计算起来是很快的。
-
-哈希函数学习的两个阶段：
-
-1. Projection Stage（dimension reduction）
-2. Quantization Stage
-
-贡献：
-
-- Isotropic Hash
-
-思想：学习一个正交阵（幻灯片21页），其目的是让大于某一阈值的feature的重要程度是一样的。
-
-- Supervised Hashing with Latent Factor Models
-- Supervised Multimodal Hashing with SCM
-- Multiple-Bit Quantization
-
-
-
-###Distributed Learning
-主讲人：我
-
-主要内容：
-
-1. 文章：Coupled Group Lasso for Web-Scale CTR Prediction
-2. 文章：Distributed Power-Law Graph Computing
-
-####文章1
-为了解决在线广告的CTR（click through rate）预测，即当某广告展示给某用户时，它被该用户点击的概率，通常的方法是LR（logistic regression），即逻辑回归。但LR的一个短板是，因其是线性的，所以无法将用户与广告之间某些微妙的非线性关系纳入。注意LR中，正则项若为2范数平方，称为标准逻辑回归；正则项若为1范数，问题通常被叫做Lasso。所以需要一种新的方法。
-
-这里的贡献是：
-
-1. CGL的似然定义中，可以纳入用户与广告之间的某些非线性关系的考量。
-2. 正则项改为参数的2-1范数，目的是是用户特征向量参数W、广告特征向量参数V中更多的行为0，行为0说明该行对应的feature没作用，即达到删除冗余feature的作用。
-
-分布式实现。这个算法具有较好的扩展性，一个master，若干slave，类似于并行计算，从而实现分布式。
-
-####文章2
-GP（graph partitioning）图分割的方法有两种：边分割；点分割。点分割在分布式计算中的通信成本会比图分割小，原因在于在不同的machine上，点分割只需保留点的copy，而边分割需要同时保留点与边的copy。
-
-切割degree大的点，即邻居多的点可以降低通信成本。
-
-
-
-
-###Stochastic Learning
-主讲人：浩锋
-
-思想：在需要用到所有节点上的信息时，通信代价往往很大，这时可以随机的选取某一个节点上的信息（比如梯度）作为替代品。
-
-
-
-###资料
-【1】[幻灯片](http://cs.nju.edu.cn/lwj/slides/BigLearning.pdf)
-
-【2】[Coupled Group Lasso for
-Web-Scale CTR Prediction in Display Advertising](http://jmlr.csail.mit.edu/proceedings/papers/v32/yan14.pdf)
-
-【3】[Distributed Power-law Graph Computing:
-Theoretical and Empirical Analysis](http://papers.nips.cc/paper/5396-distributed-power-law-graph-computing-theoretical-and-empirical-analysis.pdf)
-
-
-
diff --git "a/_posts/\347\247\221\347\240\224/2015-03-23-Newton-QuasiNewton-Method.md" "b/_posts/\347\247\221\347\240\224/2015-03-23-Newton-QuasiNewton-Method.md"
deleted file mode 100755
index a3839fa9f..000000000
--- "a/_posts/\347\247\221\347\240\224/2015-03-23-Newton-QuasiNewton-Method.md"
+++ /dev/null
@@ -1,99 +0,0 @@
----
-layout: post
-title: 牛顿法与拟牛顿法（DFP BFGS LBFGS VLBFGS）
-category: 科研
-tags: Math Optimization
-keywords: 拟牛顿法
-description:
----
-
-最近做LBFGS的并行，顺便把牛顿法、拟牛顿法顺理一下。
-
-拟牛顿法是求解非线性优化问题最有效的方法之一。考虑无约束的极小化问题
 "\min\limits_x f(x)")
，假设
 "f(x)")
为凸函数，且二阶连续可导。
-
-###原始牛顿法
-基本思想：在现有极小点估计值的附近对f(x)进行二阶泰勒展开，进而找到下一个极小点的估计值
-
-
^{-1}g_{k},&space;k=0,1,\cdots "x_{k+1}=x_{k}-(H_k)^{-1}g_{k}, k=0,1,\cdots")
-
-牛顿法具有二次收敛性，但当目标函数非二次型时，牛顿法不能保证函数稳定地下降（缺点）。
-
-
-###阻尼牛顿法
-每次迭代前需要沿迭代方向
^{-1}g_k "d_k=-(H_k)^{-1}g_k")
做线搜索，寻求最优的步长因子
，即
-
-
 "\lambda_k = \arg\min\limits_{\lambda} f(x_k+\lambda d_k)")
-
-
-###拟牛顿法
-基本思想：不使用二阶偏导数而构造出可以近似Hession或Hession的逆的正定对称阵，在“拟牛顿”的条件下优化目标函数。
-
-先推导拟牛顿条件：在
附近对
做泰勒展开，取二阶近似项
-
-
=f(x_{k+1})+\nabla&space;f(x_{k+1})(x-x_{k+1})+\frac{1}{2}(x-x_{k+1})^T\nabla^2&space;f(x_{k+1})(x-x_{k+1}) "f(x)=f(x_{k+1})+\nabla f(x_{k+1})(x-x_{k+1})+\frac{1}{2}(x-x_{k+1})^T\nabla^2 f(x_{k+1})(x-x_{k+1})")
-推出
-
-
\approx&space;\nabla&space;f(x_{k+1})+H_{k+1}(x-x_{k+1}) "\nabla f(x)\approx \nabla f(x_{k+1})+H_{k+1}(x-x_{k+1})")
-取
，推出
-
-
 "g_{k+1}-g_{k}\approx H_{k+1}(x_{k+1}-x_k)")
-
-引入记号
， 推出
-
-
(`拟牛顿条件`)
-
-它迭代过程中的hession矩阵做约束，因此，对hession对近似的
，以及对hession的逆做近似的
，可以将
-
-
或
作为指导。
-
-
-
-####DFP算法（Davidon–Fletcher–Powell formula）
-
-核心：通过迭代的方法，对hession的逆做近似。迭代格式为
-
-
（通常
）
-
-猜想
待定为
（具有对称性）
-
-
\mathbf{u}+(\beta\mathbf{v}^{\mathrm{T}}y_k)\mathbf{v} "\Rightarrow s_k=D_ky_k+\alpha\mathbf{u}\mathbf{u}^{\mathrm{T}}y_k+\beta\mathbf{v}\mathbf{v}^{\mathrm{T}}y_k=D_ky_k+(\alpha\mathbf{u}^{\mathrm{T}}y_k)\mathbf{u}+(\beta\mathbf{v}^{\mathrm{T}}y_k)\mathbf{v}")
-
-括号中是数值，将其分别简单赋值为1，-1，即
-
-
-
-其中向量u,v仍有待确定，由上面
-
-
（要此式成立，不妨直接取
）
-
-
-
-至此，校正矩阵
就已经构造出来了
-
-
-
-
-####BFGS算法（Broyden–Fletcher–Goldfarb–Shanno algorithm）
-核心公式的推导过程与DFP完全类似，只是互换了其中s{k}和y{k}的位置。BFGS直接逼近Hession矩阵B_k。(公式敲起来太累了，请自行推导)
-
-
-####LBFGS算法(limited-memory BFGS)
-不再存储完整的D_k，而是存储计算过程中的向量序列{s}，{y}。当需要矩阵D_k时，利用向量序列的计算来代替。并且，向量序列也不是全部存储，而是固定存最新的m个。
-
-若要实现并行，需要同时在x与梯度（影响y的计算）那儿求一致平均。
-
-
-###资料
-
-【1】[DFP算法](http://en.wikipedia.org/wiki/Davidon%E2%80%93Fletcher%E2%80%93Powell_formula)
-
-【2】[BFGS算法](http://en.wikipedia.org/wiki/Broyden%E2%80%93Fletcher%E2%80%93Goldfarb%E2%80%93Shanno_algorithm)
-
-【3】[LBFGS算法](http://en.wikipedia.org/wiki/Limited-memory_BFGS)
-
-【4】[Large-scale L-BFGS using MapReduce](http://papers.nips.cc/paper/5333-large-scale-l-bfgs-using-mapreduce.pdf)
-
-
-
-
-
diff --git "a/_posts/\347\247\221\347\240\224/2015-09-12-Intuition-of-Eigen-Value.md" "b/_posts/\347\247\221\347\240\224/2015-09-12-Intuition-of-Eigen-Value.md"
deleted file mode 100755
index d72d68d07..000000000
--- "a/_posts/\347\247\221\347\240\224/2015-09-12-Intuition-of-Eigen-Value.md"
+++ /dev/null
@@ -1,44 +0,0 @@
----
-layout: post
-title: 从二次型最优化问题中理解矩阵特征值的意义
-category: 科研
-tags: Math Optimization
-keywords: 特征值
-description:
----
-
-###惯例开场故事
-在某次从实验室去往食堂的路上，曾发生这样一段对话：
-
-『大师兄，为什么你对算法的理解总是那么透彻呢？为什么我很难看出它背后的思想？』
-
-『因为你去理解一个算法的时候，不能只是看懂它的形，还要去思考它的神啊~』
-
-这就是我天分不够当不了科学家的佐证吧T。T
-
-###从二次型最优化来理解
-最小化二次型目标函数
=x^TAx "f(x)=x^TAx")
，其中A为已知的实对称二阶矩阵，
，
.这个问题的求解很简单，这里以此为例来说明该问题与矩阵特征值的关系。
-
-首先，可以得到目标函数的网格图与等高线图如下。
-
-
-
-
-对矩阵A进行特征分解可以得到其特征向量为[-0.7071, 0.7071], [0.7071, 0.7071]，对应的特征值分别是0.5, 1.5.
-
-观察函数的等高线图可以知道，等高线最密集的地方，函数值变化最快，而这个函数值变化最快的方向归一化后就是[0.7071, 0.7071]，这恰好是矩阵的一个特征向量。同样地，可以观察，等高线最稀疏的地方，函数值变化最慢，变化方向则是矩阵的另一个特征向量。可以看出，`矩阵特征值的大小与函数值的变化快慢有关`，较大特征值所对应的特征向量方向上函数值的变化较快，较小特征值所对应的特征向量方向上函数值的变化较慢。
-
-进一步，对于实对称矩阵，我们总是可以对其进行相似变化，得到一个以该矩阵特征值为对角线元素的对角阵。
，其中，P为正交矩阵，有性质P的逆等于P的转置。把目标函数改写为
=x^{\textrm{T}}PBP^{\textrm{T}}x "f(x)=x^{\textrm{T}}PBP^{\textrm{T}}x")
，其中
. 相似变换的作用可以理解为将等高线图进行旋转，于是得到下面经过旋转后的等高线图。
-
-
-
-在这张图上说明矩阵特征值的意义。当函数值取1时所对应的椭圆等高线的长轴长度为
， 即由矩阵特征值0.5决定。同理，该椭圆短轴长度为
，由矩阵特征值1.5决定。
-
-二阶矩阵的理解较为直观。高阶矩阵的道理是一样的。
-
-###资料
-
-【1】[如何理解矩阵特征值](http://www.zhihu.com/question/21874816/answer/19592526)
-
-
-
diff --git "a/_posts/\347\273\230\347\224\273/2011-09-21-Cat.md" "b/_posts/\347\273\230\347\224\273/2011-09-21-Cat.md"
deleted file mode 100755
index 164ee273c..000000000
--- "a/_posts/\347\273\230\347\224\273/2011-09-21-Cat.md"
+++ /dev/null
@@ -1,11 +0,0 @@
----
-layout: post
-title: 一只猫
-category: 绘画
-tags:
-keywords:
-description:
----
-
-
-
diff --git "a/_posts/\347\273\230\347\224\273/2014-08-01-My-Paintings.md" "b/_posts/\347\273\230\347\224\273/2014-08-01-My-Paintings.md"
deleted file mode 100755
index a03b5f747..000000000
--- "a/_posts/\347\273\230\347\224\273/2014-08-01-My-Paintings.md"
+++ /dev/null
@@ -1,11 +0,0 @@
----
-layout: post
-title: 与亚敏走在石像路
-category: 绘画
-tags:
-keywords:
-description:
----
-
-
-
diff --git "a/_posts/\347\273\230\347\224\273/2014-08-03-Little-Lady .md" "b/_posts/\347\273\230\347\224\273/2014-08-03-Little-Lady .md"
deleted file mode 100755
index 18a82a796..000000000
--- "a/_posts/\347\273\230\347\224\273/2014-08-03-Little-Lady .md"
+++ /dev/null
@@ -1,11 +0,0 @@
----
-layout: post
-title: 小礼物Little Lady
-category: 绘画
-tags:
-keywords:
-description:
----
-
-
-
diff --git "a/_posts/\347\273\230\347\224\273/2014-08-04-Postmeridian-Track.md" "b/_posts/\347\273\230\347\224\273/2014-08-04-Postmeridian-Track.md"
deleted file mode 100755
index fad7b323d..000000000
--- "a/_posts/\347\273\230\347\224\273/2014-08-04-Postmeridian-Track.md"
+++ /dev/null
@@ -1,11 +0,0 @@
----
-layout: post
-title: 午后的操场
-category: 绘画
-tags:
-keywords:
-description:
----
-
-
-
diff --git "a/_posts/\347\273\230\347\224\273/2014-09-09-Mind-Eraser-Vodka.md" "b/_posts/\347\273\230\347\224\273/2014-09-09-Mind-Eraser-Vodka.md"
deleted file mode 100755
index 2ffa878c5..000000000
--- "a/_posts/\347\273\230\347\224\273/2014-09-09-Mind-Eraser-Vodka.md"
+++ /dev/null
@@ -1,11 +0,0 @@
----
-layout: post
-title: 坐在露天酒吧
-category: 绘画
-tags:
-keywords:
-description:
----
-
-
-
diff --git "a/_posts/\347\273\230\347\224\273/2014-11-03-Coast.md" "b/_posts/\347\273\230\347\224\273/2014-11-03-Coast.md"
deleted file mode 100755
index 495c8f37e..000000000
--- "a/_posts/\347\273\230\347\224\273/2014-11-03-Coast.md"
+++ /dev/null
@@ -1,17 +0,0 @@
----
-layout: post
-title: 海岸公路
-category: 绘画
-tags:
-keywords:
-description:
----
-
-
-
-###故事
-
-我生日的时候，这货送我一套绘画工具作为礼物，条件是她生日的时候我要画幅画给她。
-
-所以，生日快乐啊邓怡。
-
diff --git "a/_posts/\347\273\230\347\224\273/2015-07-05-Never-Trust-the-Word-from-Qipa.md" "b/_posts/\347\273\230\347\224\273/2015-07-05-Never-Trust-the-Word-from-Qipa.md"
deleted file mode 100755
index 87657ba44..000000000
--- "a/_posts/\347\273\230\347\224\273/2015-07-05-Never-Trust-the-Word-from-Qipa.md"
+++ /dev/null
@@ -1,21 +0,0 @@
----
-layout: post
-title: 租房奇遇记
-category: 绘画
-tags:
-keywords:
-description:
----
-
-
-
-###故事
-“人生真的是一场很长的修炼，我还处于小白阶段。”记第一次租房被耍。
-
-2015.09.29补充：
-我确实是一个很幸运的人，在被奇怪的房东耍过之后，居然又开启一段神奇的际遇：在四位善良的『陌生同学』的帮助下，我在北京没有交一分钱房租而度过了大半个暑假。真的，谢谢大家。
-
-
-
-
-
diff --git "a/_posts/\347\274\226\347\250\213/2015-10-26-Do-with-your-guitar.md" "b/_posts/\347\274\226\347\250\213/2015-10-26-Do-with-your-guitar.md"
new file mode 100755
index 000000000..f8cd8d428
--- /dev/null
+++ "b/_posts/\347\274\226\347\250\213/2015-10-26-Do-with-your-guitar.md"
@@ -0,0 +1,8 @@
+---
+layout: post
+title:
+category: 编程
+tags:
+keywords:
+description:
+---
diff --git "a/_posts/\350\257\273\344\271\246/2010-01-01-Book-List-2010.md" "b/_posts/\350\257\273\344\271\246/2010-01-01-Book-List-2010.md"
deleted file mode 100755
index 46fbfc294..000000000
--- "a/_posts/\350\257\273\344\271\246/2010-01-01-Book-List-2010.md"
+++ /dev/null
@@ -1,23 +0,0 @@
----
-layout: book
-title: 2010年阅读书单
-category: 读书
-keywords: 阅读,书单,2010
-books:
- - title: 追风筝的人
- status: 已读
- author: 卡勒德·胡赛尼
- publisher: 上海人民出版社
- language: 中文
- link: http://book.douban.com/subject/1770782/
- cover: http://img3.douban.com/lpic/s1727290.jpg
- description:
- - title: 小王子
- status: 已读
- author: 圣埃克苏佩里
- publisher: 人民文学出版社
- language: 中文
- link: http://book.douban.com/subject/1084336/
- cover: http://img5.douban.com/lpic/s1237549.jpg
- description: 这辈子都不会忘记自己手抄了一本英文小王子并画上了所有插图。
----
diff --git "a/_posts/\350\257\273\344\271\246/2011-01-01-Book-List-2011.md" "b/_posts/\350\257\273\344\271\246/2011-01-01-Book-List-2011.md"
deleted file mode 100755
index 7d460104f..000000000
--- "a/_posts/\350\257\273\344\271\246/2011-01-01-Book-List-2011.md"
+++ /dev/null
@@ -1,135 +0,0 @@
----
-layout: book
-title: 2011年阅读书单
-category: 读书
-keywords: 阅读,书单,2011
-books:
- - title: 巨流河
- status: 已读
- author: 齐邦媛
- publisher: 天下远见出版股份有限公司
- language: 繁体中文
- link: http://book.douban.com/subject/3822618/
- cover: http://img5.douban.com/lpic/s6363979.jpg
- description: 力荐。
- - title: 独唱团（第一辑）
- status: 已读
- author: 韩寒 等
- publisher: 书海出版社
- language: 中文
- link: http://book.douban.com/subject/4886245/
- cover: http://img3.douban.com/lpic/s4436571.jpg
- description:
- - title: 兄弟（上）
- status: 已读
- author: 余华
- publisher: 麦田出版
- language: 繁体中文
- link: http://book.douban.com/subject/1401425/
- cover: http://img3.douban.com/lpic/s1427312.jpg
- description: 历史上最让人想不通的：1、奥斯维辛集中营大屠杀。2、南京大屠杀。3、苏联肃清。4、文化大革命
- - title: 兄弟（下）
- status: 已读
- author: 余华
- publisher: 麦田出版
- language: 繁体中文
- link: http://book.douban.com/subject/1766678/
- cover: http://img5.douban.com/lpic/s1637979.jpg
- description:
- - title: 沉思录
- status: 已读
- author: 马可·奥勒留
- publisher: 中央编译出版社
- language: 中文
- link: http://book.douban.com/subject/2359003/
- cover: http://img3.douban.com/lpic/s2825021.jpg
- description:
- - title: 定位
- status: 已读
- author: 艾·里斯 / 杰克·特劳特
- publisher: 中国财政经济出版社
- language: 中文
- link: http://book.douban.com/subject/1017180/
- cover: http://img3.douban.com/lpic/s1081174.jpg
- description:
- - title: 那些年我们一起追的女孩
- status: 已读
- author: 九把刀
- publisher: 花山文艺出版社
- language: 中文
- link: http://book.douban.com/subject/1958227/
- cover: http://img3.douban.com/lpic/s1982025.jpg
- description:
- - title: 天使的孩子
- status: 已读
- author: 李昱宏
- publisher: 田园城市
- language: 繁体中文
- link: http://book.douban.com/subject/4207195/
- cover: http://img3.douban.com/lpic/s4454883.jpg
- description:
- - title: 转山
- status: 已读
- author: 谢旺琳
- publisher: 远流出版事业股份出版社
- language: 繁体中文
- link: http://book.douban.com/subject/2969117/
- cover: http://img5.douban.com/lpic/s2872388.jpg
- description: 期待拍成电影。这本书也是我在20岁那年骑行国道318的间接原因。
- - title: 楼下的房客
- status: 已读
- author: 九把刀
- publisher: 盖亚文化出版社
- language: 繁体中文
- link: http://book.douban.com/subject/1351669/
- cover: http://img3.douban.com/lpic/s1386454.jpg
- description:
- - title: 送你一颗子弹
- status: 已读
- author: 刘瑜
- publisher: 上海三联出版社
- language: 中文
- link: http://book.douban.com/subject/4238362/
- cover: http://img5.douban.com/lpic/s4243447.jpg
- description:
- - title: 莎拉的钥匙
- status: 已读
- author: 塔季雅娜·德·罗斯奈
- publisher: 新星出版社
- language: 中文
- link: http://book.douban.com/subject/4753689/
- cover: http://img5.douban.com/lpic/s6193906.jpg
- description:
- - title: 在漫长的旅途中
- status: 已读
- author: 星野道夫
- publisher: 世纪文景 上海人民出版社
- language: 中文
- link: http://book.douban.com/subject/3765620/
- cover: http://img3.douban.com/lpic/s4279052.jpg
- description:
- - title: 瞬间的背后
- status: 已读
- author: 乔·麦克纳利
- publisher: 人民邮电出版社
- language: 中文
- link: http://book.douban.com/subject/3544984/
- cover: http://img3.douban.com/lpic/s3613011.jpg
- description: 就是因为这本，其他的摄影书都显得苍白了点。力荐。这就是曾经沧海。
- - title: 天才在左疯子在右
- status: 已读
- author: 高铭
- publisher: 武汉大学出版社
- language: 中文
- link: http://book.douban.com/subject/4242172/
- cover: http://img5.douban.com/lpic/s6340977.jpg
- description:
- - title: 日月
- status: 已读
- author: 马家辉
- publisher: 中国人民大学出版
- language: 中文
- link: http://book.douban.com/subject/3918036/
- cover: http://img3.douban.com/lpic/s3997292.jpg
- description:
----
diff --git "a/_posts/\350\257\273\344\271\246/2012-01-01-Book-List-2012.md" "b/_posts/\350\257\273\344\271\246/2012-01-01-Book-List-2012.md"
deleted file mode 100755
index d8da1cfe7..000000000
--- "a/_posts/\350\257\273\344\271\246/2012-01-01-Book-List-2012.md"
+++ /dev/null
@@ -1,111 +0,0 @@
----
-layout: book
-title: 2012年阅读书单
-category: 读书
-keywords: 阅读,书单,2012
-books:
- - title: 上帝掷骰子吗？
- status: 已读
- author: 曹天元
- publisher: 辽宁教育出版社
- language: 中文
- link: http://book.douban.com/subject/6434486/
- cover: http://img3.douban.com/lpic/s6475450.jpg
- description:
- - title: 三体1：地球往事
- status: 已读
- author: 刘慈欣
- publisher: 重庆出版社
- language: 中文
- link: http://book.douban.com/subject/2567698/
- cover: http://img3.douban.com/lpic/s2865028.jpg
- description:
- - title: 三体2：黑暗森林
- status: 已读
- author: 刘慈欣
- publisher: 重庆出版社
- language: 中文
- link: http://book.douban.com/subject/3066477/
- cover: http://img3.douban.com/lpic/s3078482.jpg
- description:
- - title: 三体3：死神永生
- status: 已读
- author: 刘慈欣
- publisher: 重庆出版社
- language: 中文
- link: http://book.douban.com/subject/5363767/
- cover: http://img3.douban.com/lpic/s26012674.jpg
- description: 如果有一天我真的从事高科技行业，一定有这本书的功劳。
- - title: 守夜
- status: 已读
- author: 北岛
- publisher: 牛津大学出版社
- language: 繁体中文
- link: http://book.douban.com/subject/3890089/
- cover: http://img3.douban.com/lpic/s4662869.jpg
- description: 第一次见到一本诗集是作者自选，内容品质之高可见一斑。这本书，没有买到，遗憾好久。
- - title: 城门开
- status: 已读
- author: 北岛
- publisher: 牛津大学出版社
- language: 繁体中文
- link: http://book.douban.com/subject/4908991/
- cover: http://img3.douban.com/lpic/s4417111.jpg
- description:
- - title: 美国纽约摄影学院摄影教材（上册）
- status: 已读
- author: 美国纽约摄影学院
- publisher: 中国摄影出版社
- language: 中文
- link: http://book.douban.com/subject/3990022/
- cover: http://img5.douban.com/lpic/s3962469.jpg
- description: 力荐。
- - title: 美国纽约摄影学院摄影教材（下册）
- status: 已读
- author: 美国纽约摄影学院
- publisher: 中国摄影出版社
- language: 中文
- link: http://book.douban.com/subject/3990023/
- cover: http://img5.douban.com/lpic/s3962462.jpg
- description: 力荐。
- - title: 遥远的救世主
- status: 已读
- author: 豆豆
- publisher: 作家出版社
- language: 中文
- link: http://book.douban.com/subject/1322455/
- cover: http://img5.douban.com/lpic/s1332484.jpg
- description: 龚老师的推荐。
- - title: 孤独六讲
- status: 已读
- author: 蒋勋
- publisher: 广西师范大学出版社
- language: 中文
- link: http://book.douban.com/subject/4124727/
- cover: http://img5.douban.com/lpic/s4075572.jpg
- description: 突闪的念头拉着老妈去新华书店，我看上这本，老妈看上林徽因作品集。事实上我更希望老妈能认真读这本书，当她先撕裂固有思维再重组后，我就是受益最大的人。
- - title: 燕园草木
- status: 已读
- author: 许智宏 / 顾红雅
- publisher: 北京大学出版社
- language: 中文
- link: http://book.douban.com/subject/6725303/
- cover: http://img5.douban.com/lpic/s6776514.jpg
- description: 伟伦兄送的一本北大植物志。
- - title: 离春天只有二十公分的雪兔
- status: 已读
- author: 李娟
- publisher: 本事文化
- language: 繁体中文
- link: http://book.douban.com/subject/6819253/
- cover: http://img5.douban.com/lpic/s6955918.jpg
- description: 从台湾带回来打算送人的书籍，但竟因为看过之后再也舍不得送出去，至今脑海中还能浮现出那只瘦骨嶙峋的雪兔模样。
- - title: 我们仨
- status: 已读
- author: 杨绛
- publisher: 生活·读书·新知三联书店
- language: 中文
- link: http://book.douban.com/subject/1052212/
- cover: http://img3.douban.com/lpic/s1039180.jpg
- description: 一生回忆起来也就是一本书，期间平淡真情时间流逝到走在人生边上那天，会以什么样的形式浮现脑海，现在的我不得而知。
----
diff --git "a/_posts/\350\257\273\344\271\246/2013-01-01-Book-List-2013.md" "b/_posts/\350\257\273\344\271\246/2013-01-01-Book-List-2013.md"
deleted file mode 100755
index b2bff5359..000000000
--- "a/_posts/\350\257\273\344\271\246/2013-01-01-Book-List-2013.md"
+++ /dev/null
@@ -1,79 +0,0 @@
----
-layout: book
-title: 2013年阅读书单
-category: 读书
-keywords: 阅读,书单,2013
-books:
- - title: Linear System Theory and Design
- status: 已读
- author: Chen / Chi-Tsong
- publisher: Oxford University Press
- language: English
- link: http://book.douban.com/subject/20522338/
- cover: http://img3.douban.com/lpic/s26281950.jpg
- description: 自控领域教材中编写十分流畅自然的一本，也有难度。
- - title: ArcGIS Engine开发从入门到精通
- status: 已读
- author: 邱洪钢 / 张青莲 / 陆绍强
- publisher: 人民邮电出版社
- language: 中文
- link: http://book.douban.com/subject/4883293/
- cover: http://img3.douban.com/lpic/s6180500.jpg
- description: 做毕业设计需要而阅读这本书，但不敢恭维。可见技术人员的语文素养不够好的话还是写不出优秀的书籍的。
- - title: C++程序设计
- status: 已读
- author: 谭浩强
- publisher: 清华大学出版社
- language: 中文
- link: http://book.douban.com/subject/1174290/
- cover: http://img5.douban.com/lpic/s1126580.jpg
- description: 做毕业设计时复习C++而阅读，非常通俗易懂。
- - title: 看见
- status: 已读
- author: 柴静
- publisher: 广西师范大学出版社
- language: 中文
- link: http://book.douban.com/subject/20427187/
- cover: http://img5.douban.com/lpic/s24468373.jpg
- description:
- - title: The Lady
- status: 已读
- author: Barbara Victor
- publisher: Faber & Faber
- language: English
- link: http://book.douban.com/subject/2332712/
- cover: http://img5.douban.com/lpic/s2788549.jpg
- description: 对昂山素季的最初印象是中学时代在某份杂志上阅读过关于她的专题，这本书则是从她出生写到现在，令人肃严起敬又十分动容的女人。
- - title: 父与子全集
- status: 已读
- author: E.O.Plauen
- publisher: 中国工人出版社
- language: 中文
- link: http://book.douban.com/subject/1002898/
- cover: http://img5.douban.com/lpic/s1094608.jpg
- description: 看到的全是父爱。
- - title: 香蕉哲学
- status: 已读
- author: 杨昌溢
- publisher: 重庆出版社
- language: 中文
- link: http://book.douban.com/subject/20366457/
- cover: http://img5.douban.com/lpic/s24229515.jpg
- description: 因为微博平台而认识飞机的坏品味，而看了这本书。有精彩的句子，整书虽略显矫情，但也许也是作者真性情流露。
- - title: 平凡的世界
- status: 已读
- author: 路遥
- publisher: 北京十月文艺出版社
- language: 中文
- link: http://book.douban.com/subject/4237548/
- cover: http://img5.douban.com/lpic/s4231678.jpg
- description: 毕业旅行在东北和内蒙走完一圈的途中看完，故事平凡而震动人心，让人觉得所有的苦难也都并不算什么。
- - title: 乡村教师
- status: 已读
- author: 刘慈欣
- publisher: 长江文艺出版社
- language: 中文
- link: http://book.douban.com/subject/10767124/
- cover: http://img5.douban.com/lpic/s24493339.jpg
- description: 恐怕也只有大刘可以在普通的背景下让读者感受到整个宇宙的时空与爱。
----
diff --git "a/_posts/\350\257\273\344\271\246/2014-01-01-Book-List-2014.md" "b/_posts/\350\257\273\344\271\246/2014-01-01-Book-List-2014.md"
deleted file mode 100755
index 3a9b928a0..000000000
--- "a/_posts/\350\257\273\344\271\246/2014-01-01-Book-List-2014.md"
+++ /dev/null
@@ -1,135 +0,0 @@
----
-layout: book
-title: 2014年阅读书单
-category: 读书
-keywords: 阅读,书单,2014
-books:
- - title: Strategic Management text and cases
- status: 在读
- author: Dess / Lumpkin / Eisner
- publisher: McGraw-Hill Higher Education
- language: English
- link: http://book.douban.com/subject/11655278/
- cover: http://ec4.images-amazon.com/images/I/51RiZA16PkL.jpg
- description:
- - title: Convex Optimization
- status: 已读
- author: Stephen Boyd / Lieven Vandenberghe
- publisher: Cambridge University Press
- language: English
- link: http://book.douban.com/subject/1888111/
- cover: http://img5.douban.com/lpic/s24961408.jpg
- description:
- - title: 数据挖掘导论
- status: 已读
- author: Pang-Ning Tan / Michael Steinbach / Vipin Kumar
- publisher: 人民邮电出版社
- language: 中文
- link: http://book.douban.com/subject/5377669/
- cover: http://img5.douban.com/lpic/s4548758.jpg
- description:
- - title: 失控
- status: 已读
- author: Kevin Kelly
- publisher: 新星出版社
- language: 中文
- link: http://book.douban.com/subject/5375620/
- cover: http://img5.douban.com/lpic/s4554820.jpg
- description: 力荐。一本极具浪漫主义情怀的科技书，在分布式思想方面也给人启发。
- - title: 红楼梦脂汇本
- status: 已读
- author: 曹雪芹
- publisher: 岳麓书社
- language: 中文
- link: http://book.douban.com/subject/6116769/
- cover: http://img3.douban.com/lpic/s7664203.jpg
- description:
- - title: 曾文正公嘉言钞
- status: 已读
- author: 曾国藩 著/梁启超 辑
- publisher: 中国书店出版社
- language: 中文
- link: http://book.douban.com/subject/10734881/
- cover: http://img3.douban.com/lpic/s26572260.jpg
- description:
- - title: 黄金时代
- status: 已读
- author: 王小波
- publisher: 陕西师范大学出版社
- language: 中文
- link: http://book.douban.com/subject/1082406/
- cover: http://img5.douban.com/lpic/s4083149.jpg
- description:
- - title: 一桩事先张扬的凶杀案
- status: 已读
- author: 加西亚·马尔克斯
- publisher: 南海出版社
- language: 中文
- link: http://book.douban.com/subject/24382569/
- cover: http://img5.douban.com/lpic/s26546959.jpg
- description:
- - title: 三十六大
- status: 已读
- author: 冯唐
- publisher: 浙江文艺出版社
- language: 中文
- link: http://book.douban.com/subject/20278799/
- cover: http://img5.douban.com/lpic/s24430317.jpg
- description: “春水初生，春林初盛，春风十里，不如你。”
- - title: 尘埃落定
- status: 已读
- author: 阿来
- publisher: 人民文学出版社
- language: 中文
- link: http://book.douban.com/subject/1083003/
- cover: http://img5.douban.com/lpic/s1095536.jpg
- description: 前所未见的风格，有点儿魔幻现实的感觉。最近读书越来越真实的体会到：果然是哪怕知道再多的道理，人依然过不好一生。不过没关系，一生都过好了也会失去很多人生体验。比如这种明明知道思维终结的那天什么都不再有意义也仍然挣扎着想更多地填满自己的感觉，还是挺奇妙的。
- - title: 人生
- status: 已读
- author: 路遥
- publisher: 北京十月文艺出版社
- language: 中文
- link: http://book.douban.com/subject/3803820/
- cover: http://img3.douban.com/lpic/s3847911.jpg
- description:
- - title: 百年孤独
- status: 已读
- author: 加西亚·马尔克斯
- publisher: 南海出版公司
- language: 中文
- link: http://book.douban.com/subject/6082808/
- cover: http://img3.douban.com/lpic/s6384944.jpg
- description: 居然没有想象中和听说的那样晦涩难懂，4天时间，如果不是必要的睡眠和事物根本停不下来。我觉得，这本书真的写得好好。2014.5.14补：也许从现在开始当一个金鱼上校也是一件幸福的事。2014.5.26补：不是当一个金鱼上校，是清楚知道这样的状态，不刻意去做，也不刻意避免它，就跟着生活走。
- - title: 沉默的大多数
- status: 已读
- author: 王小波
- publisher: 上海三联书店
- language: 中文
- link: http://book.douban.com/subject/2986026/
- cover: http://img3.douban.com/lpic/s3006611.jpg
- description:
- - title: 牧羊少年奇幻之旅
- status: 已读
- author: 保罗·柯艾略
- publisher: 南海出版公司
- language: 中文
- link: http://book.douban.com/subject/3608208/
- cover: http://img5.douban.com/lpic/s3668327.jpg
- description:
- - title: 平如美棠
- status: 已读
- author: 饶平如
- publisher: 广西师范大学出版社
- language: 中文
- link: http://book.douban.com/subject/23008971/
- cover: http://img5.douban.com/lpic/s26835759.jpg
- description:
- - title: The Man Who Walked Between The Towers
- status: 已读
- author: Mordicai Gerstein
- publisher: Roaring Brook Press
- language: English
- link: http://book.douban.com/subject/1831015/
- cover: http://img5.douban.com/lpic/s4669638.jpg
- description:
----
diff --git "a/_posts/\350\257\273\344\271\246/2014-07-24-dream-of-red-chamber.md" "b/_posts/\350\257\273\344\271\246/2014-07-24-dream-of-red-chamber.md"
deleted file mode 100755
index e66bbc827..000000000
--- "a/_posts/\350\257\273\344\271\246/2014-07-24-dream-of-red-chamber.md"
+++ /dev/null
@@ -1,17 +0,0 @@
----
-layout: post
-title: 读《红楼梦》
-category: 读书
-tags: 随感
-keywords:
-description:
----
-
-两个月前在亚马逊上买了岳麓书社的**《红楼梦脂汇本》**，被师兄们开玩笑说太奢侈。我真想不出还有什么事比买书更朴素。小时候几番试图阅读《红楼梦》，每次都因人物关系过于复杂而放弃。说来很感谢蒋勋的细说红楼音频，尽管后来觉得他有点儿啰嗦。
-
-几天前，饭桌上谈及红楼梦，被问到最喜欢的人物是谁，一时间回答不上来。我觉得，不管在书中或者在现实里，当你尝试过体会每个角色在其长度为一生的时间轴上所经历的荣与辱、尊与卑、乐与痛、宠与屈之后，是很难去说喜欢谁或者不喜欢谁的。站在一个更高维的空间去看，每个人都过得不容易，而所有事物的发展也逃不过“分久必合，合久必分”八个大字。
-
-另外想讲讲蒋勋在他的书里一直强调的深情。人类的许多深情都在细节处体现，雨天里为你把蓑衣捋整齐的手，“我的五脏都碎了，你还只是哭”，诸如此类，能够体会的人自然能够体会。以前想不明白的事，也渐渐释然：因为有些问题，本身无解就是解。而这又有什么关系呢。
-
-
-
diff --git "a/_posts/\350\257\273\344\271\246/2014-08-25-Golden-Line.md" "b/_posts/\350\257\273\344\271\246/2014-08-25-Golden-Line.md"
deleted file mode 100755
index 5deaa9a2a..000000000
--- "a/_posts/\350\257\273\344\271\246/2014-08-25-Golden-Line.md"
+++ /dev/null
@@ -1,20 +0,0 @@
----
-layout: post
-title: 每个人的金线
-category: 读书
-tags: 随感
-keywords: 阅读，金线
-description:
----
-### 文学的金线
-冯唐在《三十六大》里试图说明文学有“金线”。和音乐、美食、普世价值等等一样，尽管标准难以量化，优劣的判断也多半掺杂了人的主观，但文学始终是有一条金线存在的，达到了就是达到了，没达到就是没达到。“文章千古事，得失寸心知”。这条金线和文章的销量没有正相关的关系，在某些时代，甚至负相关。但这条文学的金线，两三千年来香火不绝地流传了下来，掌握在少数人手里。并且，冯唐认为，他就能够明眼看出这一条文学的金线。
-
-### 手机的金线
-王自如在对锤子手机T1的测评末尾，花费占总视频1/5的时间，强调如果抛开一切情怀的包装去看产品，平心而论，这部手机与罗永浩自己宣称的东半球最好用还相差甚远。虽然如此，仍有很多人会把自己想了但没做的情感寄托到这个想了也做了的人身上。如果活在罗永浩的“情怀”之内，那么所有关于产品的技术性的好与坏、对与错都会变得界限模糊。我反感用情怀去包装产品，也反感罗永浩这个人，私以为乔布斯的iPhone在金线之上，罗永浩的锤子在金线之下。但那群活在罗永浩情怀里的人，恐怕不会赞同。
-
-### 生命的金线
-有的人真的达到了，在我认同的金线之上；以同一条金线为准，有的人就是达不到，可对我你而言的达不到，就真的是没有达到吗？我希望我和冯唐的观点一致，但事实上我并不这么认为。**云在青天，水在瓶，时光澄澈**。如果说真理就在万物的自然常态中，那么人各有态，每个人都有自己笃定的金线，每种人的存在本身就是真理。人类用认知却获得自己看到的那条金线。而认知，却构筑在好奇心、生存环境、教育等诸多因素之上。各式各样的生命形态遍布地球，生命形态的金线，又是否存在？
-
-梁启超对曾国藩的评价颇高，说曾文正公做到了“立德、立功、立言三并不朽”，文森特·梵高希望所有人都能看到他的画并因此感受到他的内心，不惜因此疯掉；阿来在《尘埃落定》里塑造的傻子，因为傻，反而能对大事做出超乎常人的正确决断，加西亚·马尔克斯的《百年孤独》中经历战争而变了个人的年老上校将所有的生命热力用于制作金鱼；痞子在酒肉中嫖赌，高僧鹤形仙眉隐于白云深处，或宏伟、或浪漫、甚至疯狂、不堪，这些生命形态，真的可以用一条金线去分割吗？不然。他们仅仅是以不同的形式存在而已，云在青天水在瓶，野猪在林，家猪在圈，蝼蚁在泥土，它们都在存在的金线之上。
-
-万紫千红，各花入各眼。南北西东，各方自有人。既然存在着，虽不欲划分生命形态的三六九等，人活着的不同阶段也有不同的追求，情怀会变。但我想，每个人在自己老了之后，心里是会有条金线的，或准之以勤，或以诚，或以快乐，或以勇敢，或成就，或物质。回顾一生，达到或者没达到，各位看官，茶余饭后，不妨斟酌斟酌。
\ No newline at end of file
diff --git "a/_posts/\350\257\273\344\271\246/2015-01-29-Book-List-2015.md" "b/_posts/\350\257\273\344\271\246/2015-01-29-Book-List-2015.md"
deleted file mode 100755
index d610530c2..000000000
--- "a/_posts/\350\257\273\344\271\246/2015-01-29-Book-List-2015.md"
+++ /dev/null
@@ -1,93 +0,0 @@
----
-layout: book
-title: 2015年阅读书单
-category: 读书
-keywords: 阅读,书单,2015
-books:
- - title: 计算广告
- status: 在读
- author: 刘鹏 王超
- publisher: 人民邮电出版社
- language: 中文
- link: http://book.douban.com/subject/26596778/
- cover: http://img3.douban.com/lpic/s28280564.jpg
- description:
- - title: Convergence Rate of Distributed Averaging Dynamics and Optimization in Networks
- status: 在读
- author: Angelia Nedic
- publisher: Now Publishers Inc (May 29, 2015)
- language: English
- link: http://www.nowpublishers.com/article/Details/SYS-004
- cover: http://7xnp1z.com1.z0.glb.clouddn.com/CRDADON.jpg
- description:
- - title: Game Theory for Applied Economists
- status: 在读
- author: Robert Gibbons
- publisher: Princeton University Press
- language: English
- link: http://book.douban.com/subject/1908414/
- cover: http://img3.douban.com/lpic/s7410801.jpg
- description:
- - title: C++ Primer
- status: 在读
- author: Stanley B. Lippman / Josée Lajoie / Barbara E. Moo
- publisher: 电子工业出版社
- language: 中文
- link: http://book.douban.com/subject/25708312/
- cover: http://img3.douban.com/lpic/s27294664.jpg
- description:
- - title: 通向实在之路-宇宙法则的完全指南
- status: 在读
- author: 罗杰·彭罗斯
- publisher: 湖南科学技术出版社
- language: 中文
- link: http://book.douban.com/subject/25823056/
- cover: http://img3.douban.com/lpic/s27215912.jpg
- description:
- - title: 数学之美
- status: 已读
- author: 吴军
- publisher: 人民邮电出版社
- language: 中文
- link: http://book.douban.com/subject/10750155/
- cover: http://img3.douban.com/lpic/s9114855.jpg
- description:
- - title: 量化投资策略
- status: 未读
- author: 理查德·托托里罗
- publisher: 上海交通大学出版社
- language: 中文
- link: http://book.douban.com/subject/24698948/
- cover: http://img3.douban.com/lpic/s26717153.jpg
- description:
- - title: 不能承受的生命之轻
- status: 已读
- author: 米兰·昆德拉
- publisher: 上海译文出版社
- language: 中文
- link: http://book.douban.com/subject/4920007/
- cover: http://img3.douban.com/lpic/s4421443.jpg
- description: 2011年在成都最大的书店无意中发现这本书的英文版，非常激动地买了下来。没想到这断断续续地阅读，直到2015年初才完整看完中文版。读这本书是比较眩晕而奇妙的，如果第一次读建议还是中文吧。读书所有的意义不是说看透什么本质，而是看透后更加用力的活着。又是同样的感觉，明知思维终止的那一天什么都不再有意义，但仍然想更多地填满自己。
- - title: 断舍离
- status: 已读
- author: 山下英子
- publisher: 广西科学技术出版社
- language: 中文
- link: http://book.douban.com/subject/24749465/
- cover: http://img4.douban.com/lpic/s26872396.jpg
- description:
- - title: 我与地坛
- status: 已读
- author: 史铁生
- publisher: 人民文学出版社
- language: 中文
- link: http://book.douban.com/subject/5910656/
- cover: http://img4.douban.com/lpic/s4594868.jpg
- description: 我甚至现在就能清楚地看见，一旦有一天我不得不长久地离开它，我会怎样想念它，我会怎样想念它并且梦见它，我会怎样因为不敢想念它而梦也梦不到它。宇宙以其不息的欲望将一个歌舞炼为永恒。这欲望有怎样一个人间的姓名，大可忽略不计。每个人的地坛都不一样吧，到最后，也不是我在地坛，而是地坛在我。
----
-
-
-
-
-
-
\ No newline at end of file
diff --git "a/_posts/\350\257\273\344\271\246/2016-01-01-Book-List-2016.md" "b/_posts/\350\257\273\344\271\246/2016-01-01-Book-List-2016.md"
new file mode 100755
index 000000000..cc593ac58
--- /dev/null
+++ "b/_posts/\350\257\273\344\271\246/2016-01-01-Book-List-2016.md"
@@ -0,0 +1,208 @@
+---
+layout: book
+title: 2016年阅读书单
+category: 读书
+keywords: 阅读,书单,2016
+books:
+ - title: 计算机科学导论（第2版）
+ status: 在读（纸质）
+ author: Behrouz Forouzan & Firouz Mosharaf
+ publisher: 机械工业出版社
+ language: 中文
+ link: http://book.douban.com/subject/1770782/
+ cover: http://img3.douban.com/lpic/s1727290.jpg
+ description:解惑
+ - title: 程序设计基础（第3版）
+ status: 正读（纸质）
+ author: 吴文虎 & 徐明星
+ publisher: 清华大学出版社
+ language: 中文
+ link: http://book.douban.com/subject/1084336/
+ cover: http://img5.douban.com/lpic/s1237549.jpg
+ description:
+ - title: C程序设计语言
+ status: 未读（纸质）
+ author:
+ publisher:机械工业出版社
+ language: 中文
+ link:
+ cover:
+ description:
+ - title:iOS 7 programming fundamentals
+ status: 未读（epub）
+ author:
+ publisher:
+ language: 英文
+ link:
+ cover:
+ description:
+ - title:Progranmming iOS 7
+ status: 未读（epub）
+ author:
+ publisher:
+ language:英文
+ link:
+ cover:
+ description:
+ - title:Programming with Objective-C
+ status: 未读（PDF）
+ author:Apple
+ publisher:
+ language: 英文
+ link:
+ cover:
+ description:
+ - title: iOS 9 Programming Fundamentals
+ status: 未读（epub）
+ author:
+ publisher:
+ language: 英文
+ link:
+ cover:
+ description:
+ - title: Programming iOS 9
+ status: 未读（epub）
+ author:
+ publisher:
+ language: 英文
+ link:
+ cover:
+ description:
+ - title:The Swift Programming Language
+ status: 未读（epub）
+ author:Apple
+ publisher:Apple
+ language:
+ link:
+ cover:
+ description:
+ - title: 数据结构与算法分析——c语言描述 （第二版）
+ status: 未读（PDF）
+ author:
+ publisher:机械工业出版社
+ language: 中文
+ link:
+ cover:
+ description:
+ - title:算法导论（第3版）
+ status: 未读（PDF）
+ author:
+ publisher:
+ language: 中文
+ link:
+ cover:
+ description:每周一章，实现一个算法
+ - title:美语发音秘籍
+ status: 在读（纸质）
+ author:Ann Cook
+ publisher:群言出版社
+ language: 中英文
+ link:
+ cover:
+ description:
+ - title: 经济学原理——微观经济学分册
+ status: 未读（纸质）
+ author:曼昆
+ publisher:北京大学出版社
+ language: 中文
+ link:
+ cover:
+ description:
+ - title: 经济学原理——宏观经济学分册
+ status: 未读（纸质）
+ author:曼昆
+ publisher:北京大学出版社
+ language: 中文
+ link:
+ cover:
+ description:
+ - title: 浪潮之巅
+ status: 未读（纸质）
+ author:吴军
+ publisher:人民邮电出版社
+ language: 中文
+ link:
+ cover:
+ description:
+ - title: 数学之美
+ status: 未读（纸质）
+ author:吴军
+ publisher:人民邮电出版社
+ language:
+ link:
+ cover:
+ description:
+ - title:公正
+ status: 未读（纸质）
+ author:
+ publisher:
+ language: 中文
+ link:
+ cover:
+ description:
+ - title: 死亡
+ status:未读（纸质）
+ author:
+ publisher:
+ language: 中文
+ link:
+ cover:
+ description:
+ - title: Pride and Prejudice
+ status: 未读（kindle）
+ author:
+ publisher:
+ language: 英文
+ link:
+ cover:
+ description:
+ - title:简爱
+ status: 未读（kindle）
+ author:
+ publisher:
+ language:英文
+ link:
+ cover:
+ description:
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+---
diff --git "a/_posts/\351\237\263\344\271\220/2015-10-26-Do-with-your-guitar.md" "b/_posts/\351\237\263\344\271\220/2015-10-26-Do-with-your-guitar.md"
new file mode 100755
index 000000000..1f9b16c28
--- /dev/null
+++ "b/_posts/\351\237\263\344\271\220/2015-10-26-Do-with-your-guitar.md"
@@ -0,0 +1,8 @@
+---
+layout: post
+title:
+category: 音乐
+tags:
+keywords:
+description:
+---
diff --git a/pages/about.html b/pages/about.html
index 8bc5acad6..829d7b30c 100755
--- a/pages/about.html
+++ b/pages/about.html
@@ -1,9 +1,9 @@
---
layout: page_title
title:
-titlename: 关于
+titlename: about
---
-
关于博客
+
博客
PainterLin.com是我用github、jekyll、markdown搭建起来的博客。博客名字直译过来是：画家林。在我看来，画家和黑客是相似的：他们都是创造者，就好像作曲家，建筑师，以及作家一样，他们的目的是创造某种美好的事物。选域名时想到这个也许代表了我的某种愿望。博客内容不一定与技术相关，技术只是世界上的一个小目录。
关于作者
diff --git a/public/img/1.jpg b/public/img/1.jpg
deleted file mode 100755
index 90fc780ad..000000000
Binary files a/public/img/1.jpg and /dev/null differ
diff --git a/public/img/1.png b/public/img/1.png
deleted file mode 100755
index 8676fad98..000000000
Binary files a/public/img/1.png and /dev/null differ
diff --git a/public/img/LAY.png b/public/img/LAY.png
deleted file mode 100755
index 386dbfc5b..000000000
Binary files a/public/img/LAY.png and /dev/null differ
diff --git a/public/img/Thumbs.db b/public/img/Thumbs.db
deleted file mode 100755
index 50f30e9fd..000000000
Binary files a/public/img/Thumbs.db and /dev/null differ
diff --git a/public/img/anyalin.png b/public/img/anyalin.png
deleted file mode 100755
index 199b42e9f..000000000
Binary files a/public/img/anyalin.png and /dev/null differ
diff --git a/public/img/birthday/1.jpg b/public/img/birthday/1.jpg
deleted file mode 100755
index 44d3b79e3..000000000
Binary files a/public/img/birthday/1.jpg and /dev/null differ
diff --git a/public/img/birthday/2.jpg b/public/img/birthday/2.jpg
deleted file mode 100755
index cb7bc5a80..000000000
Binary files a/public/img/birthday/2.jpg and /dev/null differ
diff --git a/public/img/birthday/3.jpg b/public/img/birthday/3.jpg
deleted file mode 100644
index cae6af765..000000000
Binary files a/public/img/birthday/3.jpg and /dev/null differ
diff --git a/public/img/days/1.jpg b/public/img/days/1.jpg
deleted file mode 100755
index ed8386970..000000000
Binary files a/public/img/days/1.jpg and /dev/null differ
diff --git a/public/img/days/1_1.jpg b/public/img/days/1_1.jpg
deleted file mode 100755
index ed8386970..000000000
Binary files a/public/img/days/1_1.jpg and /dev/null differ
diff --git a/public/img/days/2.jpg b/public/img/days/2.jpg
deleted file mode 100755
index 2408e9014..000000000
Binary files a/public/img/days/2.jpg and /dev/null differ
diff --git a/public/img/days/3.jpg b/public/img/days/3.jpg
deleted file mode 100755
index 072f7e482..000000000
Binary files a/public/img/days/3.jpg and /dev/null differ
diff --git a/public/img/days/4.jpg b/public/img/days/4.jpg
deleted file mode 100755
index 1e7a4282e..000000000
Binary files a/public/img/days/4.jpg and /dev/null differ
diff --git a/public/img/days/4_1.jpg b/public/img/days/4_1.jpg
deleted file mode 100755
index 63d9cf3aa..000000000
Binary files a/public/img/days/4_1.jpg and /dev/null differ
diff --git a/public/img/days/5.jpg b/public/img/days/5.jpg
deleted file mode 100755
index 2af76e36e..000000000
Binary files a/public/img/days/5.jpg and /dev/null differ
diff --git a/public/img/days/Thumbs.db b/public/img/days/Thumbs.db
deleted file mode 100755
index d0452b4a7..000000000
Binary files a/public/img/days/Thumbs.db and /dev/null differ
diff --git a/public/img/days/ca.jpg b/public/img/days/ca.jpg
deleted file mode 100644
index 8777d7cd1..000000000
Binary files a/public/img/days/ca.jpg and /dev/null differ
diff --git a/public/img/days/coast.jpg b/public/img/days/coast.jpg
deleted file mode 100755
index 94422ddc9..000000000
Binary files a/public/img/days/coast.jpg and /dev/null differ
diff --git a/public/img/deer3.png b/public/img/deer3.png
deleted file mode 100755
index 22d6e7fec..000000000
Binary files a/public/img/deer3.png and /dev/null differ
diff --git a/public/img/dog.jpg b/public/img/dog.jpg
deleted file mode 100755
index 5100c5637..000000000
Binary files a/public/img/dog.jpg and /dev/null differ
diff --git a/public/img/food/fish.jpg b/public/img/food/fish.jpg
deleted file mode 100755
index 72e7ed037..000000000
Binary files a/public/img/food/fish.jpg and /dev/null differ
diff --git a/public/img/food/lobster.jpg b/public/img/food/lobster.jpg
deleted file mode 100755
index 184f69232..000000000
Binary files a/public/img/food/lobster.jpg and /dev/null differ
diff --git a/public/img/food/mung.jpg b/public/img/food/mung.jpg
deleted file mode 100755
index 0ace0a197..000000000
Binary files a/public/img/food/mung.jpg and /dev/null differ
diff --git a/public/img/food/pigtrotter.jpg b/public/img/food/pigtrotter.jpg
deleted file mode 100755
index ac8c801ec..000000000
Binary files a/public/img/food/pigtrotter.jpg and /dev/null differ
diff --git a/public/img/food/steak.jpg b/public/img/food/steak.jpg
deleted file mode 100755
index 9850b7b77..000000000
Binary files a/public/img/food/steak.jpg and /dev/null differ
diff --git a/public/img/food/toast.jpg b/public/img/food/toast.jpg
deleted file mode 100755
index d9a2d2591..000000000
Binary files a/public/img/food/toast.jpg and /dev/null differ
diff --git a/public/img/latexslideseg.jpg b/public/img/latexslideseg.jpg
deleted file mode 100755
index fc32e2374..000000000
Binary files a/public/img/latexslideseg.jpg and /dev/null differ
diff --git a/public/img/linuxDIRTREE.gif b/public/img/linuxDIRTREE.gif
deleted file mode 100755
index da1195a7d..000000000
Binary files a/public/img/linuxDIRTREE.gif and /dev/null differ
diff --git a/public/img/longhair.gif b/public/img/longhair.gif
deleted file mode 100755
index 5f2942def..000000000
Binary files a/public/img/longhair.gif and /dev/null differ
diff --git a/public/img/longhair.jpg b/public/img/longhair.jpg
deleted file mode 100755
index 588741082..000000000
Binary files a/public/img/longhair.jpg and /dev/null differ
diff --git a/public/img/me.jpg b/public/img/me.jpg
new file mode 100644
index 000000000..2d294a68e
Binary files /dev/null and b/public/img/me.jpg differ
diff --git a/public/img/posts/2_1norm.png b/public/img/posts/2_1norm.png
deleted file mode 100644
index a8e0d0025..000000000
Binary files a/public/img/posts/2_1norm.png and /dev/null differ
diff --git a/public/img/posts/contour_of_f.png b/public/img/posts/contour_of_f.png
deleted file mode 100644
index 4a00e269a..000000000
Binary files a/public/img/posts/contour_of_f.png and /dev/null differ
diff --git a/public/img/posts/contour_of_f_B.png b/public/img/posts/contour_of_f_B.png
deleted file mode 100644
index b55989a5b..000000000
Binary files a/public/img/posts/contour_of_f_B.png and /dev/null differ
diff --git a/public/img/posts/jobs.png b/public/img/posts/jobs.png
deleted file mode 100644
index 0889b4b74..000000000
Binary files a/public/img/posts/jobs.png and /dev/null differ
diff --git a/public/img/posts/matrix_norms.JPG b/public/img/posts/matrix_norms.JPG
deleted file mode 100755
index 0b6fab5cd..000000000
Binary files a/public/img/posts/matrix_norms.JPG and /dev/null differ
diff --git a/public/img/posts/mesh_of_f.png b/public/img/posts/mesh_of_f.png
deleted file mode 100644
index 1c8c81fe3..000000000
Binary files a/public/img/posts/mesh_of_f.png and /dev/null differ
diff --git a/public/img/posts/resume.jpg b/public/img/posts/resume.jpg
deleted file mode 100755
index e444c0485..000000000
Binary files a/public/img/posts/resume.jpg and /dev/null differ
diff --git a/public/img/posts/s1.jpg b/public/img/posts/s1.jpg
deleted file mode 100755
index d974f65ba..000000000
Binary files a/public/img/posts/s1.jpg and /dev/null differ
diff --git a/public/img/posts/s1.png b/public/img/posts/s1.png
deleted file mode 100755
index 7ab29fd9c..000000000
Binary files a/public/img/posts/s1.png and /dev/null differ
diff --git a/public/img/posts/score0.png b/public/img/posts/score0.png
deleted file mode 100755
index dee6834e9..000000000
Binary files a/public/img/posts/score0.png and /dev/null differ
diff --git a/public/img/posts/trace.JPG b/public/img/posts/trace.JPG
deleted file mode 100755
index c36c5b505..000000000
Binary files a/public/img/posts/trace.JPG and /dev/null differ
diff --git a/public/img/posts/vector_norms.JPG b/public/img/posts/vector_norms.JPG
deleted file mode 100755
index 633d5c3bd..000000000
Binary files a/public/img/posts/vector_norms.JPG and /dev/null differ
diff --git a/public/img/shorthair.jpg b/public/img/shorthair.jpg
deleted file mode 100755
index 57dcbe954..000000000
Binary files a/public/img/shorthair.jpg and /dev/null differ
diff --git a/public/img/shorthair1.jpg b/public/img/shorthair1.jpg
deleted file mode 100755
index adefcfaf1..000000000
Binary files a/public/img/shorthair1.jpg and /dev/null differ
diff --git a/public/img/skim.png b/public/img/skim.png
deleted file mode 100755
index 47856d690..000000000
Binary files a/public/img/skim.png and /dev/null differ
diff --git a/public/img/slide_eg.JPG b/public/img/slide_eg.JPG
deleted file mode 100755
index 9873de979..000000000
Binary files a/public/img/slide_eg.JPG and /dev/null differ
diff --git a/public/img/tao.jpg b/public/img/tao.jpg
new file mode 100644
index 000000000..317d8b6bb
Binary files /dev/null and b/public/img/tao.jpg differ
diff --git a/public/img/work.jpg b/public/img/work.jpg
deleted file mode 100755
index f648dfe70..000000000
Binary files a/public/img/work.jpg and /dev/null differ
diff --git a/public/img/yellowdeer.jpg b/public/img/yellowdeer.jpg
deleted file mode 100755
index 18328a684..000000000
Binary files a/public/img/yellowdeer.jpg and /dev/null differ