Machine learning is one of the trending technology in the IT field which Everyone is trying to master. In this race Master the ML Tricks By Knowing the Foundations of Machine learning.
What Actually Is ML? How It's Different From AI.
Machine Learning is a Subdomain of Artificial Intelligence, so before getting to know about ML we need to have a good understanding of AI. First, let's talking how a programmer writes a program, first, he sees the problem and writes a problem that can solve it. What if the problems modify into something else or there are more to the problem which the programmer dint notice. In this cases he should again write the program, for example, if you are playing a chess game for the first time without knowing the rules you just do some random moves and then learn about the moves and game, then on further practicing, again and again, he will master the game. Same way when a program or machine learns the problem and develops the code according to the problem it is said " The machine is learning". If you are new to AI go through our last article here What's So Trendy About AI That Everyone Went Crazy Over It? where we talked about AI more precisely.
The most popular definition of AI is " … the science of making machines do things that would require intelligence if done by men" by Marvin Minsky. The first-ever program with AI was Strachey's checkers (draughts) program which was used to play checker games with real players. Later on, programs for Chess were also developed now we see smart bots which can play more sophisticatedly than a real-world player all due to the self-learning capability of the machine.
So Machine learning and Deep learning are like extended uses of AI. So what about machine learning?. "... Machine learning is a program which becomes more accurate as it predicts the future without explicitly programmed to do so". In machine learning, there are different ways to train the data so it can predict future statistics by learning from old experiences. for example, let's consider an example of how humans identify mammals if the animal gives an egg it's not a mammal if the animal is cold-blooded then it's not a mammal, etc. So when we think about a mammal we think about all these attributes or features which it has. So machine learning algorithm also identifies the features of the problem which are essentials for the prediction of the future state of the problem.
Ways in which We can teach Machine to learn.
The most popular types of machine learning are:
- Supervised learning
- Unsupervised learning
- Semi-Supervised learning
- Reinforcement learning
These are the most used algorithms based on different scenarios lets see in detail about each of them.
Supervised Learning: -
When are given the data of the problem with labels, and the machine predicts the missing labels by learning from the given models then it is known as supervised learning. When are investing in a stock market we see the shares of the company during the past few days and come to a conclusion to invest in it or not. This is a simple example of supervised learning where we see the previous dataset values and try to predict the future dataset. Here there can be two types of dataset labels is Discrete values and other is Continous Values. When the answer is in the form of "yes", "no" or "1", "0" and any values which are predefined and are not infinite then it is said to be continuous. When the label values range between some number etc, for example, the age of the person is a continuous label, the Salary of a person is also a continuous variable.
So in the above example, a new input is given to the model then the learning algorithm finds the relevant label for the input by seeing the pre-existing dataset in the data. There are a variety of supervised learning models that are divided based on the above-mentioned labels.
- Classification: It is used when the Labels are discrete values. For example, whether a team will win or not, whether the price will win or not, whether your salary will be higher $100k or not etc.
- Regression: This is used when the Labels are continuous values. For example, what is the age of a person, What will be your salary after 10 years, How many goals a team will make? etc.
Unsupervised Learning: -
As in supervised learning, we give all the responses or datasets, but in unsupervised learning, there will be no dataset from which the machine learns. In this, the algorithms find the relations between the data and categorize them accordingly. First, the algorithm inferences the inputs and finds similarities between them, then it makes clusters of the similar datasets which are unlabeled. As the labels are not provided here we cannot find the accuracy of the algorithm. The most commonly used unsupervised learning algorithm is clustering analysis. Clustering analysis is a statistical method of grouping or categorizing the data based upon the similarities.
If we are given a group of items let's consider this item like fruit so the first thing which comes to our mind to classify them based on the characteristics of the fruit like color, size, etc. which is nothing but a kind of unsupervised learning where the machine was able to categorize the items and keep the future items incorrect cluster, etc. Most of the Development projects start from unsupervised learning then to supervised learning.
Semi-Supervised Learning: -
Semi-supervised learning comes in between supervised and unsupervised learning. In this, the algorithm is the dataset but only a few are labeled rest are all re unlabeled data. This is done when we have to predict a large amount of data but only a few examples of the data are given. So it uses both supervised learning and unsupervised learning techniques, First, it learns necessary information from the labeled data then it categorizes the unlabeled data then it gives labels to them based on the inference of the labeled data.
Reinforcement learning: -
Reinforcement learning also comes in between supervised and unsupervised learning. In this type of learning the algorithm knows whether the answer or label is right or wrong but doest know how it's right or wrong. It should figure out the reason for it and learn from it. It has to try all the different ways in which the dataset can be proved and find the best reason for it. Reinforcement learning imposes rewards, policies, and punishments that increase the speed of learning.
So in the above picture, we can see that the trainer trains the dog the way the dataset trains the algorithm. It dog doesn't have much intelligence to understand humans but it learns that it's doing something good if it gets a reward and something is wrong if it gets a punishment. So in the same way learning from the data based on the results and improving is called reinforcement learning. The only duty of the agent here is to maximize the rewards and gain as much data.