
ML
Xiaomin-Wu
这个作者很懒,什么都没留下…
展开
专栏收录文章
- 默认排序
- 最新发布
- 最早发布
- 最多阅读
- 最少阅读
-
Machine Learning by Andrew Ng --- Support Vector Machine
Aha, after the intricate BP neural network(I spend more than 24 hours to finish that exercise,but even i finish ,i am sure i am not understand BP neural network well),now i finally meet the powerful原创 2015-03-19 10:43:56 · 1000 阅读 · 0 评论 -
Deep Learning by Andrew Ng --- self-taught
本次UFLDL练习大致流程:通过对标记为5-9的数字图像进行self-taught特征提取(笔画特征),获得特征参数opttheta。use opttheta to obtain a(2) which represente the labeled input data.Training and testing the logistic regression model(with softmax原创 2015-04-06 21:17:29 · 1237 阅读 · 0 评论 -
Machine Learning by Andrew Ng --- neural network learning
The step of this exercise is show in the pdf which i have updoaded.Neural network of this exercise is not easy to finish,okay,let me show U.nnCostFunction:function [J grad] = nnCostFunction(原创 2015-03-14 13:50:52 · 1172 阅读 · 0 评论 -
Deep Learning by Andrew Ng --- Sparse coding
稀疏编码介绍稀疏编码算法是一种无监督学习方法,它用来寻找一组“超完备”基向量来更高效地表示样本数据。稀疏编码算法的目的就是找到一组基向量 ϕi\begin{align}\mathbf{\phi}_i\end{align} ,使得我们能将输入向量 x 表示为这些基向量的线性组合: x=∑i=1kaiϕi\begin{align}\mathbf{x} = \sum_{i=1}^k a_i \math原创 2015-04-18 16:41:29 · 1492 阅读 · 0 评论 -
Deep Learning by Andrew Ng --- stacked autoencoder
When should we use fine-tuning?It is typically used only if you have a large labeled training set; in this setting, fine-tuning can significantly improve the performance of your classifier. However, if原创 2015-04-08 20:05:04 · 2740 阅读 · 1 评论 -
Deep Learning by Andrew Ng --- Sparse Autoencoder
这是UFLDL的编程练习,因为只看到第一章节,还没有看到向量化,所以本篇博客只注意对算法的理解,没有注意向量化。因为进入机器学习领域也只有一个多月,许多错误之处望指出。 传统的神经网络一般用于监督学习(supervised learning),一般而言,SVM比传统的backpropagation分类算法更有效。但是最近神经网络在无监督学习(unsupervised learning)领域开始大放原创 2015-03-29 16:11:09 · 1437 阅读 · 0 评论 -
受限波尔茨曼机概要
本文根据论文–introduction to Restricted Boltzmann Machine(中国科技论文在线),删去了一些公式推理,作简要概括,主要为编程实现而总结。(具体内容请参考原文)受限波尔茨尼机结构及相关函数: 基于对比散度的RBM快速学习算法 上面算法中的每个for语句都可以使用向量化来加速算法。 另外,其他一些改进的学习算法: 参数设置 与一般的神经网络训练原创 2015-05-22 23:13:57 · 1614 阅读 · 0 评论 -
caffe --- 网络参数
如何在Caffe中配置每一个层的结构 —核心网络 最近刚在电脑上装好Caffe,由于神经网络中有不同的层结构,不同类型的层又有不同的参数,所有就根据Caffe官网的说明文档做了一个简单的总结。1. Vision Layers1.1转载 2015-07-14 16:40:34 · 1866 阅读 · 0 评论 -
CNN for Visual Rcognition --- Stanford 2015 (一)
总结斯坦福2015李飞飞教授以及Andrej Karpathy教授的上课的slides中关于CNN的一些内容一:神经网络实验的基本策略:1.对输入数据进行预处理: 2.网络结构和数据集设置:(1)-随机初始化weights为一些比较小的数(fan-in,fan-out),bias设置为0(2)-利用好cv集,可以在少量epoch的结果下选择最好的params,然后进行更多的epoch3.分析实验结原创 2015-08-11 11:47:16 · 3068 阅读 · 4 评论 -
CNN for Visual Rcognition --- Stanford 2015 (二)
接着上一篇的内容:四:Transfer Learning:1.对于数据量少或者中等的情况,迁移学习很有用2.基于ImageNet的实验,将ImageNet的所有类的各分一半为A,B:(1).先训练A部分,然后将前n层的参数保存好;再重新初始化后n+层的参数,用B部分训练;再将前面保存好的参数,和后面训练B部分得到的参数结合,在B的验证集上进行验证:(2).先训练A部分,训练完A后重新初始化n+层后面原创 2015-08-13 21:45:56 · 1624 阅读 · 0 评论 -
浅析SAE与DBM(Deep Learning)
笔者实力有限,诸多错误之处还望指出。 SAE与DBN两个都可以用于提取输入集特征的算法。SAE是由多个Spase AutoEncoder堆叠而成,单个Spase AutoEncoder的结构如下: 在堆叠成SAE时的结构如下: 以上SAE的结构可以化分为两个sparse autoencoder和一个 softmax(这里不讨论softmax).其中的两个sparse autoenc原创 2015-09-08 19:51:54 · 7461 阅读 · 0 评论 -
GBDT--简单理解
梳理1.Model Ensemble 可以分为三大类:Bagging,Boosting, Stacking.2.Boosting可以说是一个思想(框架),而Adaboost等算法只是其一个子类,记得ICCV2015有一个结合CNN和Boosting的工作获得了Best Paper Award?:3.Boosting的 前向分布算法(在每一步求解弱分类器Φ(m)和其参数w(m)的时候不去修改之前已经求原创 2016-03-17 19:36:46 · 3304 阅读 · 0 评论 -
Deep learning by Andrew Ng --- Linear Decoder
Sparse Autoencoder Recap:Because we used a sigmoid activation function for f(z(3)), we needed to constrain or scale the inputs to be in the range [0,1], since the sigmoid function outputs numbers in th原创 2015-04-10 16:31:37 · 1364 阅读 · 0 评论 -
Deep Learning by Andrew Ng --- Softmax regression
这是UFLDL的编程练习。Weight decay(Softmax 回归有一个不寻常的特点:它有一个“冗余”的参数集)后的cost function和梯度函数:cost function:J(θ)=−1m⎡⎣∑i=1m∑j=1k1{y(i)=j}logeθTjx(i)∑kl=1eθTlx(i)⎤⎦+λ2∑i=1k∑j=0nθ2ij\begin{align}J(\theta) = - \frac{原创 2015-04-04 15:52:16 · 1751 阅读 · 0 评论 -
Machine Learning by Andrew Ng --- Anomaly Detection and Recommender systems
In anomaly detection problems,every features corresponding to one formula of Gaussian distribution.To start anomaly detection,the steps you should do are the fellows:Using all the examples of each fe原创 2015-03-26 10:35:59 · 1189 阅读 · 0 评论 -
Machine Learning by Andrew Ng---Linear Regression with multiple variables
Just as last bog,input Octave commands show in the figure below: There are something different from last example with one variable,We should add a step namedFeature Normalization.using原创 2015-03-08 16:11:16 · 671 阅读 · 0 评论 -
Machine Learning by Andrew Ng --- Logistic Regression by using Regularization
When we face a problem like this :To create a classifier of this dataset,we may add some features by using Feature mapping.But,to avoid overfiting, we also should use Regularization :原创 2015-03-08 17:32:04 · 698 阅读 · 0 评论 -
机器学习常见算法分类汇总
机器学习无疑是当前数据分析领域的一个热点内容。很多人在平时的工作中都或多或少会用到机器学习的算法。这里IT经理网为您总结一下常见的机器学习算法,以供您在工作和学习中参考。机器学习的算法很多。很多时候困惑人们都是,很多算法是一类算法,而有些算法又是从其他算法中延伸出来的。这里,我们从两个方面来给大家介绍,第一个方面是学习的方式,第二个方面是算法的类似性。学习方式根据数据类型的转载 2015-03-27 16:28:33 · 1100 阅读 · 0 评论 -
对话机器学习大神Yoshua Bengio
Yoshua Bengio教授(个人主页)是机器学习大神之一,尤其是在深度学习这个领域。他连同Geoff Hinton老先生以及 Yann LeCun(燕乐存)教授,缔造了2006年开始的深度学习复兴。他的研究工作主要聚焦在高级机器学习方面,致力于用其解决人工智能问题。他是仅存的几个仍然全身心投入在学术界的深度学习教授之一,好多其他教授早已投身于工业界,加入了谷歌或Facebook公司。转载 2015-03-20 16:46:52 · 1381 阅读 · 0 评论 -
Machine Learning by Andrew Ng ---Linear Regression with one variable
Linear Regression is often used for predicting .The simple step of Linear Regression is always like: step1 load your data step2 plot your data step3 use Gradien Descent to learn the原创 2015-03-08 14:48:26 · 1071 阅读 · 0 评论 -
Machine Learning by Andrew Ng --- Logistic Regression with two classes
this is a example to build a logistic regression model to predict whether a student gets admitted into a university.As usual,loading and plotting your data:and the plotData.m file is :原创 2015-03-08 16:43:14 · 904 阅读 · 0 评论 -
Machine Learning by Andrew Ng --- K-means
K-mean is a kind of cluster algorithm, It allows us to find characteristics in the No characteristic data. The two important steps of K-mean : One is that clustering every example to its right原创 2015-03-21 16:02:12 · 1120 阅读 · 0 评论 -
Machine Learing by Andrew Ng --- PCA
U can use PCA to speed up your learning algorithm,but before it ,U’d better to run your learning algorithm with no PCA first(This is What Andrew Ng said).And U can also compress data by using PCA.amazi原创 2015-03-21 16:42:54 · 969 阅读 · 0 评论 -
Machine Learning by Andrew Ng --- Logistic Regression of Multi-class Classification
We will use Logistic Regression to recognize the number 1-10.Loading data,Plotting data.as the picture below:Vectorizing regularized logistic regressionThen learn t原创 2015-03-10 19:48:43 · 924 阅读 · 0 评论 -
Deep Learning by Andrew Ng --- PCA and whitening
这是UFLDL的编程练习。具体教程参照官网。PCAPCA will find the priciple direction and the secodary direction in 2-dimention examples. then x~(i)=x(i)rot,1=uT1x(i)∈R.\begin{align}\tilde{x}^{(i)} = x_{{\rm rot},1}^{(i)}原创 2015-04-02 21:00:32 · 1119 阅读 · 0 评论 -
深度学习资料汇总
title: 深度学习资料汇总 categories: - Summarize整理了平时自己用到的一些DL的资源网站deeplearning.netUFLDL机器学习日报kaggle winner solution斯坦福人工智能实验室课程列表+ppt强化学习资料汇总机器学习资料汇总卷积网络Trick卷积网络Trick-EnglishOpenCV(python/c++)D原创 2016-03-06 15:35:42 · 879 阅读 · 0 评论