Stanford University, open class - Andrew Ng - Machine Learning - Introduction Section

introduction

1.1 Welcome

---

1.1.1 Applications

• Data Mining
• Medical Applications
• Computational Biology
• Handwriting recognition
• Natural Language Processing
• Computer Vision

1.2 Machine learning is what?

---

1.2.1 definitions

• Arthur Samuel: the case of performing specific programming, giving the field of computer learning ability
• Tom Mitchell: A program is considered from experience E learning, to solve the task T , achieve performance measure P , if and only if, with experience E later, after P evaluation , program processing T performance has improved
  • Example 1: The experience E * is thousands of times on the program self-practice experience, and task T is playing chess, performance metrics P **, that is it when some new rival game, the probability of winning the game

1.2.2 Learning Algorithm

• Unsupervised Learning: programmers will teach the computer how to complete the task
• Supervised learning: computer own learning, to complete the task
• other:
  • Reinforcement Learning
  • Recommended system

1.3 supervised learning

---

1.3.1 definitions

• The basic idea is that each sample dataset has a corresponding "right" answer, algorithm to make predictions based on these samples.

1.3.2 Use

• Whenever you want based on a given input, predict an outcome, and there is an input - output for example, should use supervised learning

1.3.3 Category

• Regression: The goal predict a continuous value, suggesting a series of continuous-valued attributes
  • Example 1: level of education, age and place of residence to predict a person's annual income, the predicted result is a monetary value
  • Example 2: Prediction Rate The area of the house
• Category: predictive classification labels that list of selectable from predefined, launched the results of a discrete set of
  • Example 1: mail systems, the identification of spam
  • Example 2: is performed according to the attributes of tumor classification of tumors, benign or malignant Analyzing
    

1.3.4 several small concepts

• Generalization: If a model can make accurate predictions on data not seen, we say that it can generalize from the training set to test set, our aim is to build a model of accuracy as high as possible generalization
• Overfitting: too much attention to the details of the training set, we get a good performance on the training set, but can not be generalized to the new data model, then there OVERFITTING
• Underfitting: model is too simple, the change may not be able to seize the entire contents of the data as well as data. Even model performance on the training set is very poor

1.3.5 Quiz

1.4 Unsupervised Learning

---

1.4.1 definitions

• No advance notice algorithm some of the information, the data without any labels or the same tag or label that is not in the data set to find some kind of structure, to find relationships between data
• It is a learning strategy, algorithm to large amounts of data, and allow us to find some kind of algorithm from the data structure

1.4.2 Examples

• News News grouped in the composition associated news
• Microscopic analysis of the DNA data of the individual clusters of different classes or different types of groups
• Computer cluster work together, more efficient set of grouped
• Analysis of social networks, automatically gives the grouping of friends
• Analysis of astronomical data
• Cocktail problem, isolate sound