Machine Learning Introduction
First of all, In general, I would have to mention that Machine learning had grown out of early work in AI(Artificial Intelligence) And over the last 20 years it’s been viewed as a sort of growing a new capability for computers, but in particular, it’s clear that there are many applications that we can’t program by hand, for example extracting the characters from cursive handwriting or flying a helicopter are impossible by explicit coding. The successful approach is to use a learning algorithm to make computers learn by themselves.
Learning algorithm in simple word
All of us using a learning algorithm without being aware of it. for instance, every time we send an email, writing a check, using credit cards, purchase from a website, and … we are using such an algorithm to perform an action.
Machine Learning Definition
By Arthur Samuel (1959). (Informal definition) :
Field of study that gives computers the ability to learn without being explicitly programmed.
he has proved it by writing a checkers program which had learned to play much better than himself could
By Tom Mitchell (1988). (Formal definition) :
A Computer program is said to learn from experience E with respect to some task T and some performance measure P, if it’s performance on T, as measured by P, improves with experience E.
Overview of 4 Major Concepts in Machine Learning :
1- Supervised Learning
In this approach, we are supervising the algorithm by giving the right answer to it to learn the association between input and output with an existing dataset to make predictions for new data.
let’s check an example. I want to check the price of my house. Firstly I need to prepare a dataset from the existing house price and put a straight line to it.
then let’s find my house size on the x-axis and predict the price as done below :
Supervised learning is good at regression and classification problems.
The term regression refers to the fact that the variable you’re trying to predict is a continuous value and price.
In the classification problem, the variable you’re trying to predict is discrete rather than continuous.
let’s go through an example. we are going to learn the algorithm to predict whether or not a certain tumor is malignant. This can be done by the size of the tumor. The variable you’re trying to predict is a discreet value. It’s either zero or 1 in the Y-axis, and so your data set may look something like this :
What should we do to use more than one variable to try to predict?
In the case of patients of different ages and who have different tumor sizes, we are going to predict whether a tumor is malignant or benign where a plus indicates a malignant tumor, and an “O” indicates a benign tumor.
2- Learning Theory
The main concept of the theory is how to take the tools of machine learning and really apply them well to create a proper learning algorithm.
Unlike supervised learning, we will not get any right answer in the dataset and just should find out an interesting structure through a dataset.
one of the general examples, in this case, would be in the market segment. we need to recognize a structure in such datasets by clustering.
This type of learning refers to problems where you don’t do one-shot decision-making. It means that you are usually asked to make a sequence of decisions over time.
The basic idea behind a reinforcement learning algorithm is called a reward function just like trying to train a dog. So every time your dog does something good, you say, “Good dog,” and you reward the dog and do it in the opposite way as well. So over time, your dog will learn to do the right things to get more of the positive rewards.
It’s exactly done the same for teaching a helicopter to fly, and of course up to the learning algorithm to figure out how to maximize the good reward signals and minimize the bad rewards signals.
Finally, it’s clear that reinforcement learning is applied to many problems in robotics.