Machine Learning (1)

今天复习一下上学时学习的机器学习的概念。

1. 一些机器学习的应用场景

   1. 手写识别
   2. 物体识别
   3. 情感识别（I love this course 是积极的，I will take another course 是消极的）
   4. 遥感信息图像分类（通过俯视图查看区域种类）
   5. 网页搜索
   6. 语音识别
   7. 垃圾邮件识别
   8. 机器人
   9. 医疗健康
   10. 指纹识别
   11. 面部识别
   12. 自动驾驶
   13. 卫星图树种类分类识别

   一些免费的数据集

   1. Image/Video Databases (comprehensive): https://homepages.inf.ed.ac.uk/rbf/CVonline/Imagedbase.htm
   2. http://archive.ics.uci.edu/
   3. handwritten digits: http://yann.lecun.com/exdb/mnist/
   4. https://image-net.org/

2. 一些术语

   1. 分类
   2. 回归
   3. 聚类
   4. 标签 Label
   5. 特征 Features
   6. 损失函数 Loss Function
   7. 消耗函数 Cost Function
   8. 准确度 Accuracy
   9. 单个数据 Example / Sample
   10. 训练集 Labeled Sample
   11. 测试集 Unlabeled Sample
   12. 模型 Model
   13. 算法 Algorithms

3. 一些例子
   
   分类
   
   物体分类与识别
   
   训练集和测试集

4. 机器学习的分类
   监督学习 —— Labelled data
   无监督学习　——　Labelled & Unlabeled Data
   强化学习　——　Reward　System

监督学习
● Supervised Learning - Classification
○ Features: image/pixels
○ Label: Cat and Dog
● Task: Given a picture of car or dog, predict its label.

● Supervised Learning - Regression
○ Features: size of house
○ Label: house price
● Task: Given a house’s size, predict the selling price.

Recap:
● A model trained on historical data that are
labeled or known ground truth (e.g.
previous house sales example).
● Once the model is trained, it can then be
tested on new data to predict the label.

无监督学习
● What if No ground truth available?
● Unsupervised Learning - Clustering
○ Features: Length and Width for types
of flowers.
○ Label: No Label for unsupervised!
● Task: Cluster together the data into
similar groups/patterns.

强化学习
Reinforcement learning works through trial and error which actions yield the greatest
rewards.

深度学习
A family of machine learning methods that uses deep
architectures to learn high-level feature representations.

Parametric and non- Parametric models
• Parametric models:
• Have a fixed number of parameters
• Faster to use, simpler
• Stronger assumptions on the data distribution
• e.g., Logistic regression

• Non- Parametric Models:
• The number of parameters grow with the amount of
training data
• Flexible on data distribution
• Slower, risk of overfitting, more data requirement
• e.g., K-nearest neighbors (KNN)

监督学习工作流程

那么如何评价一个模型呢？

Supervised Learning —— Classification
Evaluation Metrics
Accuracy , Recall, Precision, Confusion matrix

Accuracy: Correctly Classified divided by
total samples.

Confusion Matrix: Shows the actual and predicted labels from a
classification problem.

Recall: the proportion of actual positives was
identified correctly.

Precision: the proportion of positive identifications
was actually correct.