1. Convolutional Neural Networks_Computer vision

1.1. 컴퓨터 비전 소개

Convolutional Networks(콘볼루션망)

컴퓨터 비전은 빠르게 발전 - 딥러닝

자율주행 자동차 - 주변 어디에 다른 자동차나 보행자 감지
안면 인식 - 예전보다 훨씬 잘 작동
ex) 핸드폰 얼굴 인식. 집에 잠긴 문 인식
핸드폰의 음식 사진이나 호텔 사진, 혹은 재미있는 풍경사진들을 인식 하는 app
새로운 유형의 예술을 탄생

1.1.1. 컴퓨터 비전 분야의 빠른 발전이 완전히 새로운 어플리케이션 생산

1.1.2. 컴퓨터 비전 시스템 자체를 만드는 것 못한다 해도, 컴퓨터 비전을 연구하는 단체가 그 동안 아주 창의적이고 아주 독창적으로 새로운 신경망 구조와 알고리즘을 이용해 일해 왔기 때문에 실제로 다른 부냐로도 교차 적용할 수 있도록 북돋워 준다는 것


1.2. Computer Vision Problems

1.2.1. Image Classification
Cat 구분

1.2.2. Object detection
자율 자동차에 물체 인식

1.2.3. Neural Style Transfer(신경 스타일 이동)
기존 사진을 다른 스타일의 이미지로 결합

 기존 강의에서 64 * 64 이미지에 3 rgb 채널을 사용하여 12288 feature가 됩니다.

하지만 64 * 64 이미지는 상당이 작은 이미지

만약 더 큰 이미지들로 작업을 하면

1000 * 100 * 3 = 300만입니다.

300만의 입력feature를 가지고 있다면 이는 곧 300만 차원일 것이라는 것을 의미합니다.

결국, 첫 번째 은닉 층에서 아마도 1000 은닉 층이 있을 것이고 그럼 총 무게가 W1매트릭스라는 것입니다.

만일 여러분이 기본 혹은 연결된 네트워크를 사용한다면, 이 매트릭스는 1000 x 3000000 차원의 매트릭스가 될 것입니다.

왜냐하면 X는 300만을 곱한 R이 되기 때문이죠. 3m, 300만 표시하기 위해 m을 사용하고 있습니다.
 이 매트릭스가 매우 매우 거대한 30억 파라미터에 이른다는 것이죠.

그리고 이 많은 파라미터로 신경망 네트워크가 과대 적합을 방지하도록 충분한 데이터를 얻는 것을 어렵습니다.

또한, 30억 파라미터를 가진 신경망을 단련하는 것에 대한 컴퓨터 요구사항과 메모리 요구사항은 실행불가능 합니다.

더 나은 회선 작업을 구현 해야 컨볼루션 신경망의 근본적 구성 요소 중의 하나이기도 하니까요.
edge detection의 예시를 통해 컨볼루션을 설명하도록 하겠습니다.

What is Reinforcement Learning?

Environment => State, Reward





Agent =>Action

005. Machine learning W1 05 Model Representation

Linear regression with one variable
Model representation

Supervised Learning
Given the "right answer" for each example in the data.

Regression Problem
Predict real-valued output


Training set




Learning Algorithm







Size of house ->       h         -> Estimated price

x              hypothesis                 (estimated value of y)

h maps from x's to y's


How do we represent h?

Linear regression with one variable.
Univariate linear regression.


Linear regression with one variable
Cost function

004. Machinelearing W1 04UnsupervisedLearning

Unsupervised Learning

- Organize computing clusters
- Social network analysis
- Market segmentation
- Astronomical data analysis


Cocktail party problem

Speaker1                    Microphone1




Speaker2                     Microphone2


[W,s,v] = svd((repmat(sum(x.*x,1),size(x,1),1).*x)*x');


Of the following examples, which would you address using an unsupervised learning algorithm?(Check all that apply.)

- Given email labeled as spam/not spam, learn a spam filter.

- Given a set of news articles found on the web, group them into set of articles about the same story.


- Give a database of customer data, automatically discover market segments and group customers into different market segments.

- Give a dataset of patients diagnosed as either having diabetes or not, learn to classify new patients as having diabetes or not.

003. Machine learning _W1_03_Supervised Learning

Introduction Supervised Learning


Housing price prediction.

Supervised Learning
"right answers" given



Regression: Predict continuous
valued output(price)


Breast cancer(malignant, benign)








Classification
Discrete valued output(0 or 1)

- Clump Thickness
- Uniformity of Cell Size
- Uniformity of Cell Shape
...

You're running a company, and you want to develop learning algorithms to address each of two problems.

Problem 1: You have a large inventory of identical items. You want to predict how many of these items will sell over the next 3 months.

Problem 2: You'd like software to examine individual customer accounts, and for each account decide if it has been hacked/compromised.

Should you treat these as classification or as regression problems?

a) Treat both as classification problems.

b) Treat problem 1 as a classification problem, problem 2 as a regression problem.

c) Treat problem 1 as a regression problem, problem 2 as a classification problem.  T

d) Treat both as regression problems.

001. Machine learning 01_Welcome

Machine Learning
 - Grew out of work in AI
 - New capability for computers

Example:
 - Database mining
    Large datasets from growth of automation/web.
    E.g., Web click data, medical records, biology, engineering

 - Applications can't program by hand.
    E.g., Autonomous helicopter, handwriting recognition, most of Natural Language Processing(NLP), Computer Vision.

 - Self-customizing programs
    E.g., Amazon, Netflix product recommendations

 - Understanding human learning(brain, real AI).


Machine Learning definition
- Arthur Samuel(1959). Machine Learning: Field of study that gives computers the ability to learn without being explicitly programmed.

- Tom Mitchell(1998) Well-posed Learning Problem: A computer program is said to learn from experience E with respect to some task T and some performance measure P, if its performance on T, as measured by P, improves with experience E.


Suppose your email program watches which emails you do or do not mark as spam, and based on that learns how to better filter spam. What is the task T in this setting?

a) Classifying emails as spam or not spam.              T

b) Watching you label emails as spam or not spam.         E

c) The number(or Fraction) of emails correctly classified as spam/not spam.    P

d) None of the above-this is not a machine learning problem.     



Machine learning algorithms:

-  Supervised learning
-  Unsupervised learning


Others: Reinforcement learning, recommender systems.

Also talk about: Practical advice for applying learning algorithms.

WEB SERVICE

- Service delivered over the web?

What aout interoperability?

Will a .Net application be able to call the services in the JAR?

WEB SERVICE -W3C DEFINITION

Software system designed to support interoperable machine-to-machine interaction over a network.

3KEYS
- Designed for machine-to-machine(or application-to-application) interaction

- Should b interoperable- Not platform dependent

- Should allow communication over a network