Introduction to Machine learning

Chapter 1 Introduction

1.1 What are machine learning?

T o Solve a problem on a computer, we need an algorithm. An algorithm was a sequence of instructions that should was carried out to transform the input to output. For example, one can devise a algorithm for sorting. The input is a set of numbers and the output is their ordered list. For the same task, there is various algorithms and we may be interested in finding the most efficient one, requiring t He least number of instructions or memory or both.

F Or some tasks, however, we don't have a algorithm-for example, to-tell spam emails from legitimate email. We know what the input Is:an e-mail document, the simplest case is a file of characters. We know what the output should be:a yes/no output indicating whether the message was spam or not. We don't know how to transform the input to the output. What can is considered spam changes in time and from individual to individual.

W Hat we lack in knowledge, we do up to in data. We can easily compile thousands of example messages some of which we know to being spam and what we want are to "learn" what C Onstitutes spam from them. In other words, we would like the computer (machine) to extract automatically the algorithm for this task. There is no need to learn to sort numbers, we already has algorithms for that; But there is many applications for which we don't have a algorithm but does have example data.

With advances in computer technology, we currently has the ability to store and process large amounts of data, as well as To access it from physically distant locations over a computer network. Most data acquisition devices is digital now and record reliable data. Think, for example, of a supermarket chain that have hundreds of stores all over a country selling thousands of goods to MI Llions of customers. The point of Sale terminals record the details of each transactions:date, customer identification code, goods bought and Their amount, total money spent, and so forth. This typically amounts to gigabytes of data every day. What is the supermarket chain wants it to being able to predict who is the likely customers for a product. Again, the algorithm for this are not evident; It changes in time and by geographic location. The stored data becomes useful only if it is analyzed and turned into information so we can make use of, for example, To make predictions.

W E May is able to identify the process completely, but we believe we can construct a good and useful approximation. That approximation is explain everything, but may still is able to account for some part of the data. We believe that thought identifying the complete process may is possible, we can still detect certain or patterns Larities. This was the niche of machine learning. Such patterns may help us understand the process, or we can use those patterns to make predictions:assuming that the Futu Re, at least of the near future, won't is much different from the past when the sample data is collected, the future pred Ictions can also is expected to being right.

Application of machine learning methods to large databases are called data mining. The analogy is, a large volume of each and raw material are extracted from a mine, which when processed leads to a smal L amount of very precious material; Similarly, in data mining, a large volume of data are processed to construct a simple model with valuable use, for example, Having a high predictive accuracy. Its application areas is abundant:in addition to retail, in finance banks analyze their past data to build models In credits applications, fraud detection, and the stock market.

1.2.5 Reinforcement Learning

I n Some applications, the output of the system is a sequence of action. In such a case, a single action was not important; What's important is the policy that's the sequence of correct actions to reach the goal. There is no such thing as the best action with any intermediate state; An action is good if it was part of a good policy. In such a case, the machine learning program should is able to assess the goodness of policies and learn from past good AC tion sequences to is able to generate a policy. Such learning methods is called reinforcement learning algorithms.

Chapter 2 Supervised learning

We discuss supervised learning starting from the simplest case, which are learning a class from its positive and negative E Xamples. We generalize and discuss the case of multiple classes, then regression, where the outputs is continuous.

2.1 Learning a Class from Examples

L Et us say we want to learn the class, C, of a "family car". We have a set of examples of cars, and we have a group of people so we survey to whom we show these cars. The people look at the cars and label them; The cars that they believe is family cars is positive examples, and the other cars is negative examples. Class learning is finding a description, which is shared by all positive examples. Class learning is finding a description, which is GKFX by all positive examples and none of the negative examples. Doing This, we can make a prediction:given a car that we had not seen before, by checking with the description learned, We'll be able to say whether it's a family car or not. Or we can do knowledge Extraction:this study is sponsored by a car company, and the an aim may is to understand what PEO Ple expect from a family car.

Chapter 3 Bayesian decision theory

We discuss probability theory as the framework for making decisions under uncertainty. In classification, Bayes ' rule was used to calculate the probabilities of the classes. We generalize to discuss what we can make rational decisions among multiple the actions to minimize expected risk. We also discuss learning Association rules from data.

3.1 Introduction

Programming computers to make inference from data are a cross between statistics and computer science, where statisticians Provide the mathematical framework of making inference from data and computer scientists work on the efficient Implementat Ion of the inference methods.

Data comes from a process, which is not completely known. This lack of knowledge are indicated by modeling the process as a random process. Maybe the process is actually deterministic, but because we don't have access to complete knowledge about it, we model it As random and use probability theory to analyze it. At this point, it is a good idea to jump the appendix and review basic probability theory before continuing with this Chapter.

Chapter 4 parametric Methods

have discussed how to make optimal decisions when the uncertainty are modeled using probabilities, we now see how we can Estimate these probabilities from a given training set. We start with the parametric approach for classification and regression. We discuss the semiparametric and non parametric approaches in later chapters. We introduce bias/variance dilemma and model selection methods for trading off model complexity and empirical error.

4.1 Introduction

A statistic is any value, which is calculated from a given sample. In statistical inference, we do a decision using the information provided by a sample.

Introduction to Machine learning