Machine Learning: Introduction

What is Machine Learning?

Before jumping into the topic of machine learning we have to understand what learning is. Learning is the process of acquiring skills and knowledge. We can understand it through activity. Please follow the instructions given below

1. Close your Eyes
2. Imagine you are in an empty room
3. Imagine in the middle of that room there is a tree
4. Open your eyes

Can you tell which tree it was? For most of us, it was not any specific tree rather it is a generic tree on the basis of the idea we have about trees on our heads. This idea of the tree or any other objects can be called a model. We pass any object we see through this model and this model tells us what we are seeing based on our knowledge that we have learned.

The Term Machine Learning was coined by Arthur Samuel in 1959 A.D.

According to him, Machine Learning is a field of study that gives computers the capability to learn without being explicitly programmed.

http://infolab.stanford.edu/pub/voy/museum/samuel.html

How Does it Work?

In the traditional approach, data is supplied to the algorithm (program), which processes the data and provides an output. The program contains the processing logic. This approach is very useful in things like calculations where there is an obvious step by step procedure to follow, like finding the area of a circle. When we get into things like pattern finding this approach does not work well. For example, we want to write a program to detect a face in a photo. If we program a computer to detect a face, it will get too complicated too quickly. Rather than writing the program, we can supply the images with the label on them and create a mathematical representation that will act as the program that detects the face.

Types of Machine Learning Algorithms

Supervised Learning

As the name suggests in supervised machine learning there is supervision or a teacher. In supervised machine learning, the data is already labeled meaning that the data is tagged with the correct answer. The supervised algorithm tries to build a generic model that can give the answer for unseen data points.

Classification

Classification is the process of assigning a class label to examples from the problem domain. A simple example is classifying emails as spam or not spam.

• Binary classification refers to predicting one of two classes. Example: classifying cancer or not cancer from medical image data
• Multi-class classification involves predicting one of more than two classes. Example: classifying a given handwritten digit into one of the numbers from 0 to 9

Regression

Regression is used to find the relationship between the target (called dependent variable y) and one or more predictors (called the independent variable X)

Linear Regression assumes the following model:

𝑦=𝑋𝛽+𝑐+𝜖

X: data point
𝛽: coefficients
c: intercept
𝜖: error, cannot be explained by the model
y: target

Unsupervised Learning

Unsupervised learning is a type of machine learning algorithm in which the labels are not provided in the dataset. IT finds the structure in its input. The most common unsupervised learning method is cluster analysis.

Clustering

Clustering is the process of grouping the data points into different groups based on similar features. The basic idea is that data points with similar features should be in the same group. It is basically a collection of objects on the basis of similarity and dissimilarity between them.

Reinforcement Learning

Reinforcement Learning means learning a process or pattern of behavior.

Reinforcement Learning has four essential elements

1. Agent: The model trained to complete the specified job.
2. Environment: The world (real or virtual), in which the agent performs actions.
3. Action: Any activity that can be performed by the agent in the given environment.
4. Rewards: The evaluation of an action, which can be positive or negative.

Basically, the agent is set in an environment where the action is rewarded on the basis of the previously set goal. The OPEN AI bot that played and won against the human DOTA2 players was trained using reinforcement learning where the goal is to win the game by destroying the enemies ancient, the game setup(game mechanics) is the environment and the action that a character (hero) can make is the action. The reinforcement algorithm gives higher rewards to action that will lead to winning.

In this post, the basic theoretical aspect of machine learning was discussed. In upcoming posts, I will be writing on how to get started with machine learning using Python programming language and libraries like scikit learn.