Machine Learning and Artificial intelligence work together hand in hand. To implement, AI we need machine learning to develop knowledge and accuracy. Machine learning can be defined as a knowledge and skill development/enhancement by the machine through its experience of work and then allowing it to learn on its own without being programmed. That is, the machine will perform some tasks and based on those outcomes it will learn what happens how and then explore more.

It can be thought of as similar to raising a child. The child is educated by parents and teachers, thereby enabling him/ her to decide on him/ her how to earn and live life.

In today’s article, we shall be discussing the three types of machine learning:

**1. Supervised Learning**

**2. Unsupervised Learning**

**3. Reinforcement Learning**

To get a basic gist of these, suppose of a scenario wherein there is a teacher who is teaching a student. Now, the teacher knows everything about the subject, i.e. the teacher knows the correct answers to all the questions asked in the book. The teacher will first make the student understand the lesson and then ask him/ her to solve some questions to test his knowledge. This is an example of supervised learning. Unsupervised learning will be leaving the students with question papers directly to identify what kind of questions they are and how to answer them. Reinforcement learning will be letting the student be on the earth and discover the laws of gravity, motion, math, languages, etc on his/ her own.

## 1. S**upervised Learning **

It is a learning technique such that there are a trainer and trainee AI who work together to achieve an already known goal (correct answer). The teacher/supervisor will come out with some example set of input and output pairs. These example sets are known to be as training sets that the AI system will use to further find results to the given inputs.

The data provided as input is labeled as to teach robot “This is an apple”, “This is an orange”, etc along with the visuals of how these fruits look like and then we can use our machine to count the number of apples and oranges.

It will solve regression and classification problems. Regression problems are the ones related to continuous quantity, for instance, to find out the species of a flower over a continent. Classification problems are the concrete division between a given quantity, for say, business contacts and family contacts in your phone directory.

**Algorithms followed in supervised learning: **

**Regression Algorithms**

Both linear and logistic regression algorithms are used for regression techniques. These algorithms mark the relationship between dependent and independent variables and show us how a change in the independent variable results in a change in the dependent variables.

**Support Vector Machines**

Capable of performing both linear and non-linear classification by plotting each data points on a maximum-margin hyperplane with a maximum margin classifier.

**Random Forest Algorithm**

Think of a forest, what do you imagine? Trees. The more the number of trees the stronger the forest. Random forest algorithm takes a training set into consideration and makes up decision trees for possible outcomes. Then on the basis of voting, the optimal solution is chosen. This algorithm is best for classification purposes.

**K Nearest Neighbour**

It is made on the belief that similar data points are nearer to each other. The algorithm calculates the distance between the nearest neighbors and then the closer data points are classified under one category or group.

## 2. **Unsupervised Learning**

In unsupervised learning the input sets are unknown, i.e. we do not have any idea of their classification, identification, etc. So, here, the datasets that are related to each other are grouped together. Then several datasets are formed and predictions are made on the basis of their similarities and differences.

In this system, the data is not labeled. You taught your machine how orange looks like and how an apple looks like. Now you bring in different species of apples & oranges and ask the machine what they are.

It is for association and clustering problems. Association problems deal with the association of similar things, for example, students who buy pencils also buy erasers and sharpeners. Clustering is more like finding people for the target market based on their interests. For say, grouping people of common interests like electronics for target marketing. Now, more electronic gadget ads will be shown to them.

**Algorithms followed in supervised learning:**

**K Means Clustering **

K Means Clustering will group similar data points together and then perform analysis to discover hidden sequences, patterns, etc. The grouping is done by finding centroids in the given dataset randomly and classifying nearest neighbors under the same group.

**Fuzzy C Means **

C Means works upon finding similarities between data points and tries to put similar data points in one group. Now, one data point can belong to more than one groups but it is pressured to create groups such that they as dissimilar as possible.

**Apriori**

Apriori is made up of the word “prior” and apriori algorithm works if prior knowledge to the datasets is available. It is mostly done for relational analysis to find our relations between two or more items.

**Self Organizing Maps**

It follows a topological ordering of data points wherein we randomly select data points. The data points are rearranged on the basis of their weights. They follow the competitive learning approach and are best for the visualization of problems. Travelling Salesman Problem is neatly solved by them.

## 3. **Reinforcement Learning**

It is actually derived from the psychological concept of reinforcement learning wherein positive and negative rewards are given to the trainees based on their performances.

In this, the AI is lured with positive points to do a task. Several systems will engage in doing the same chore separately than on the basis of who delivered the result fastest/ accurate or maybe some other parameter, positive points will be given to the machines. If the result is not desired then negative points are awarded to the machines.

There is no predefined data. In short, the machine is not exposed to anything similar to that. Like, you left a machine (a rover) on the moon and asked it to explore.

A simpler example could be a smart vacuum cleaner is left in a totally unknown environment [your home]. Now, the input will depend totally on its actions. It will depend on the movement of the vacuum cleaner that it will crash with the couch or not. Then it will gradually learn from its accidents how to find a way in your house.

**Algorithms used in reinforcement learning algorithm:**

**Markov Decision Tree**

At a given time, the process is in some state “S”. At this point, the decision-maker does some action and enters another state “S2” based on which it gets a reward. So, here, the state depends upon the decision-maker, i.e. the current state.

**Q Learning**

Quality Learning where quality is the worth of an action to provide us with future rewards. For that, it takes random actions based on the Greedy approach and finds out the optimum solution.

**Deep Q Learning**

Q Learning fails to estimate the Q values for unseen states and thus it has no idea what actions to perform when such thing happens. Deep Q learning tries to solve this problem with neural networks to estimate the Q values. It goes through convolution and fully connected layers respectively to estimate the right values.

**SARSA**

Stands for State-Action-Reward-State-Action. It is the same as Q Learning with one difference that Q learning does not follow any policy but SARSA does follow policy and is known as an on-policy algorithm.

**Deep Deterministic Policy Gradient**

Quite similar to Q learning. There is an actor who decides the actions for a specifier state-based policy parameter. There is a critic to evaluate the policy function estimated by the actor according to the temporal difference (TD) error.

**Conclusion**: So in this article we’ve learned about 3 types of machine learning and how these differ from each other based on their inputs and the environment provided to them. Then we explained various algorithms followed in these learnings.