[学习笔记] CS131 Computer Vision: Foundations and Applications:Lecture 9 深度学习
深度学习
So far this week
- Edge detection
- RANSAC
- SIFT
- K-Means
- Linear classifier
- Mean-shift
- PCA/Eigenfaces
- Image features
Current Research
- Learning hierarchical representations from data
- End-to-end learning: raw inputs to predictions
- can use a small set of simple tools to solve many problems
- has led to rapid progress on many problems
- Inspired by the brain(very loosely!)
Deep learning for different problems
vision tasks
visual recognition
object detection: what and where
object segmentation
image caption
visual question answering
super resolution
image retrieval
style transfer
outside vision tasks
- Machine Translation
- Text Synthesis
- Speech Recognition
- Speech Synthesis
Motivation
Data-driven approach:
- collect a dataset of images and labels
- use machine learning to train an image calssifier
- evaluate the classifier on a withheld set of test images
feature learning
what is feature learning?[^what is feature learning]
deep learning
Supervised learning
linear regression
neural network
neural networks with many layers
Gradient descent
how to find the best weights \(w^T\)
which way is down hill?
gradient descent
fancier rules:
- Momentum
- NAG
- Adagrad
- Adadelta
- Rmsprop
这里以后可以再 看看!
Backpropagation
a two-layer neural network in 25 lines of code
import numpy as np
D,H,N = 8, 64,32
#randomly initialize weights
W1 = np.random.randn(D,H)
W2 = np.random.randn(H,D)
for t in xrange(10000):
x = np.random.randn(N,D)
y = np.sin(x)
s = x.dot(W1)
a = np.maxium(s,0)
y_hat = a.dot(W2)
loss = 0.5*np.sum((y_hat-y)**2.0)
dy_hat = y_hat - y
dW2 = a.T.dot(W2.T)
da = dy_hat.dot(W2.T)
ds = (s > 0)*da
dW1 = x.T.dot(ds)
W1 -= learning_rate*dW1
W2 -= learning_rate*dW2
[^what is feature learning]:
In Machine Learning, feature learning or representation learningis a set of techniques that learn a feature: a transformation of raw data input to a representation that can be effectively exploited in machine learning tasks. This obviates manual feature engineering, which is otherwise necessary, and allows a machine to both learn at a specific task (using the features) and learn the features themselves.
Feature learning is motivated by the fact that machine learning tasks such as classification often require input that is mathematically and computationally convenient to process. However, real-world data such as images, video, and sensor measurement is usually complex, redundant, and highly variable. Thus, it is necessary to discover useful features or representations from raw data. Traditional hand-crafted features often require expensive human labor and often rely on expert knowledge. Also, they normally do not generalize well. This motivates the design of efficient feature learning techniques, to automate and generalize this.
Feature learning can be divided into two categories: supervised and unsupervised feature learning, analogous to these categories in machine learning generally.
In supervised feature learning, features are learned with labeled input data. Examples include Supervised Neural Networks, Multilayer Perceptron, and (supervised) dictionary Learning.
In unsupervised feature learning, features are learned with unlabeled input data. Examples include dictionary learning, independent component analysis, autoencoders, and various forms of clustering.
[学习笔记] CS131 Computer Vision: Foundations and Applications:Lecture 9 深度学习的更多相关文章
- [学习笔记] CS131 Computer Vision: Foundations and Applications:Lecture 1 课程介绍
课程大纲:http://vision.stanford.edu/teaching/cs131_fall1718/syllabus.html 课程定位: 课程交叉: what is (computer) ...
- [学习笔记] CS131 Computer Vision: Foundations and Applications:Lecture 2 颜色和数学基础
大纲 what is color? The result of interaction between physical light in the environment and our visual ...
- [学习笔记] CS131 Computer Vision: Foundations and Applications:Lecture 4 像素和滤波器
Background reading: Forsyth and Ponce, Computer Vision Chapter 7 Image sampling and quantization Typ ...
- [学习笔记] CS131 Computer Vision: Foundations and Applications:Lecture 3 线性代数初步
向量和矩阵 什么是矩阵/向量? Vectors and matrix are just collections of ordered numbers that represent something: ...
- Computer Vision: Algorithms and ApplicationsのImage processing
实在是太喜欢Richard Szeliski的这本书了.每一章节(after chapter3)都详述了该研究方向比較新的成果.还有很多很多的reference,假设你感兴趣.全然能够看那些參考论文 ...
- Deep Learning 10_深度学习UFLDL教程:Convolution and Pooling_exercise(斯坦福大学深度学习教程)
前言 理论知识:UFLDL教程和http://www.cnblogs.com/tornadomeet/archive/2013/04/09/3009830.html 实验环境:win7, matlab ...
- Sony深度学习框架 - Neural Network Console - 教程(1)- 原来深度学习可以如此简单
“什么情况!?居然不是黑色背景+白色文字的命令行.对,今天要介绍的是一个拥有白嫩的用户界面的深度学习框架.” 人工智能.神经网络.深度学习,这些概念近年已经涌入每个人的生活中,我想很多人早就按捺不住想 ...
- 百度DMLC分布式深度机器学习开源项目(简称“深盟”)上线了如xgboost(速度快效果好的Boosting模型)、CXXNET(极致的C++深度学习库)、Minerva(高效灵活的并行深度学习引擎)以及Parameter Server(一小时训练600T数据)等产品,在语音识别、OCR识别、人脸识别以及计算效率提升上发布了多个成熟产品。
百度为何开源深度机器学习平台? 有一系列领先优势的百度却选择开源其深度机器学习平台,为何交底自己的核心技术?深思之下,却是在面对业界无奈时的远见之举. 5月20日,百度在github上开源了其 ...
- Python入门学习笔记4:他人的博客及他人的学习思路
看其他人的学习笔记,可以保证自己不走弯路.并且一举两得,即学知识又学方法! 廖雪峰:https://www.liaoxuefeng.com/wiki/0014316089557264a6b348958 ...
随机推荐
- [luogu P2756 ] 飞行员配对方案问题 (最大流)
强行做裸题做了两个小时..我果然太水了QAQ 题目背景 第二次世界大战时期.. 题目描述 英国皇家空军从沦陷国征募了大量外籍飞行员.由皇家空军派出的每一架飞机都需要配备在航行技能和语言上能互相配合的2 ...
- [luogu P1962] 斐波那契数列(带快速幂矩阵乘法模板)
题目背景 大家都知道,斐波那契数列是满足如下性质的一个数列: • f(1) = 1 • f(2) = 1 • f(n) = f(n-1) + f(n-2) (n ≥ 2 且 n 为整数) 题目描述 请 ...
- Hadoop安装和使用
1.安装 1.1.下载hadoop-2.5.1.tar.gz 1.2.解压至安装目录 tar -zxv -f hadoop-2.5.1.tar.gz -C ../soft/ 1.3.配置hadoop相 ...
- 使用IDEA 中 实现springboot 热部署 (spring boot devtools版)
第一步:添加springboot的配置文件 首先我先贴出我的配置 添加依赖包 <!-- spring boot devtools 依赖包. --> <dependency> & ...
- solrj 操作 solr 单机版
一.导入 jar 包 <dependency> <groupId>org.apache.solr</groupId> <artifactId>solr- ...
- C#中的public protected internal private
1.常见的四种方位修饰符关系下图中的protected internal是并集的关系,意思是在命名空间内是internal的,在命名空间外是protected的 2.sealed final seal ...
- DynaActionForm(动态ActionForm)的使用
在struts中利用DynaActionForm(动态ActionForm)可以节省代码的编写. 1.在struts-config.xml中配置DynaActionForm:加入这个Form中有三个属 ...
- [SharePoint2010开发入门经典]一、SPS2010介绍
本章概要: 1.熟悉SPS基本特性 2.理解SPS基础架构 3.开发SPS工具
- [Android]RecyclerView的简单演示样例
去年google的IO上就展示了一个新的ListView.它就是RecyclerView. 下面是官方的说明,我英语能力有限,只是我大概这么理解:RecyclerView会比ListView更具有拓展 ...
- Bash脚本中的操作符
一.文件測试操作符 假设以下的条件成立将会返回真. -e 文件存在 -a 文件存在 这个选项的效果与-e同样. 可是它已经被"弃用"了, 而且不鼓舞使用. -f 表示这个文件是一个 ...