Scientists Say: Machine learning

Scientists Say: Machine learning is not AI, it’s just a new way to do AI.

Machine learning and Artificial Intelligence are two different things because they don’t actually do the same thing. Machine Learning is how the computer learns by itself from the data it receives and it doesn’t care for human intentions. On the other hand, Artificial Intelligence is how the computer should get its answers from itself (its own thoughts or predictions) before giving answers back to the humans.

In the past, Machine Learning was something which was almost non-existent. With time, Machine Learning became much more effective in almost all the fields of the science. Many companies have taken this advantage and started using Machine Learning in making their services and the results are stunning.

We are going to cover some of the most commonly used technologies in the data science that allow machines to learn.

One last thing, you will find very few companies that use machine learning in their services. You will also find most of them are still working at early phases of the adoption of machine learning. A lot more time will be needed to take the knowledge into practice and it will bring more cost as well. This is the reason we are focusing mainly on the technologies which can be used for free or for cheap.

Let’s talk about some most commonly used machine learning technologies.

Machine Learning: Basics

Machine learning is a type of artificial intelligence that is inspired by the brains of robots and computers. In the field of machine learning, we have four different ways to look into the nature of machine learning. We look at the theory behind the learning, classification rules, algorithms that are used by the machine learning algorithms and how are they applied.

Before going deeper into the technology, let’s start with the basics of machine learning.

A. Machine Learning Models

A machine learning model is all the mathematical and programming formula that the computer uses to figure out some answer out of the data that it gets. A model is usually implemented as a series of rules which are used by the computer to figure out an answer from the data that is provided.

Here is a basic example of how a model works.

This is called as a linear regression model. The model above is used to find the relationship between the prices of houses and the amount of bedrooms it has.

Linear regression models follow the concept of finding out the answer using a straight line. It is similar to adding the data point on a graph and finding a straight line to get the answer. In linear regression models, the model is developed in such a way that it is able to predict how much it can predict before the actual answer is given. As you can see, it basically tells the computer how good is it at predicting. This line in the chart is the line that is used to predict the price of house when it is given the amount of rooms it has.

But of course, linear regression models are not able to tell what is the relationship between any two points in the data.

Linear regression models are quite simple to understand and also easy to implement. The data points in the model needs to be linearly dependent on each other in order to work effectively.

B. Classification

Here we will talk about classification which is not a part of machine learning, but the simplest kind of machine learning. This is basically how the computer predicts something using a class by class basis. A binary classifier is mostly used in classification. A more popular and preferred method is the multiclass classifier because it works way better than the binary classifier. The data points in a classification are organized into categories; then the computer will create different kinds of rules that can figure out the class of a data point.

C. Regression

Here we will talk about regression. Regression models are a type of rule that predicts the value of dependent variables from the independent variables. We have some types of regression.

Regression techniques are very wide ranging from simple polynomial to neural networks. Polynomials are very popular and are considered to be the best among all the regression models.

A simple non-linear polynomial regression uses the following formula to find the output of the given inputs.

Here x and y are the values of the given independent variables X and Y respectively.

D. Algorithms

Here we will talk about algorithms. Algorithms work the same way as the rules of a model, but they are based on computer algorithms. There are some very common algorithms. One of them is the decision tree algorithm.

A decision tree classification algorithm will create its rules using the concepts of mathematical expressions and logical relationships.

A tree itself can be built on three different levels which is also called as three levels of analysis.

Decision tree algorithms help us to solve a problem much more easily than by directly applying the mathematical formulas.

Machine Learning Algorithms

There are lots of machine learning algorithms, each of which belongs to a group of different algorithms. This list presents some of the best-known machine learning algorithms. You can read these algorithms very well in their books. We have tried our best to explain them based on our understanding.

A. KNN (K-Nearest Neighbors)

K-Nearest Neighbor (k-NN) is one of the simplest algorithms, and it is known for its high prediction accuracy. It is a supervised learning algorithms because it needs to train the data on how the data will form during the prediction. For example, if you are building an application to find out the number of a person which has its bank card which has been lost, the user should tell the k-NN when that person lost the card so that it can calculate and tell you the number of people.

B. k-Means

In k-means clustering, each data point belongs to one cluster and gets an individual data label. Clusters are separated by k-means algorithm. k-means clustering assigns the data point to the center of the cluster.

C. Naive Bayes

Naive Bayes is a type of probabilistic classifier and it is used in spam filtering. These classifiers use statistical probabilities to predict whether the data point will be spam. To use naive bayes, you need some data, a prior, and model the data, which is the part you have a lot less information about.

D. Support Vector Machines (SVMs)

SVMs are some of the very well-known machine learning algorithms and are mostly used in classification. The concept of this algorithm is to find the boundaries between the data points.

E. Decision Trees

Decision trees are some of the easy-to-use machine learning algorithms that have a very low prediction error and are widely used in data mining and predictive modeling. If you try to apply a straight line to all the data points, you will encounter multiple points outside the prediction range and this causes a lot of noise problems.

F. Decision Trees

Decision trees or Classification trees can be used in the same area where a decision tree algorithms are used, like data mining and predictive model. Classification trees are built with decision trees like k-means clustering.

The classification tree algorithms are very easy to understand and implement because a classification tree model is much simpler than decision tree models.

In a classification tree models, each node is a decision. When we talk about the classification tree model, we mean that each node is a binary decision. These trees have two basic branches which are also called as two types of decision node, namely terminal node and non-terminal node.

G. Clustering

Clustering is basically unsupervised learning algorithm. It does not require a training phase and uses the similar features and the distance between them to form groups of data points. The algorithm is used to analyze the relationships of the data points. Commonly used clustering algorithms are K-means clustering, K-Medoids clustering and OPTICS clustering.

H. Determining Relationships

This is not a machine learning algorithm but determining relationship is something that machine learning algorithms can easily learn. This type of algorithms is based on the idea that if given data points are close to each other than it may be likely that they have some common properties. This is the way k-means clustering is used to assign the data points which have the same features in the same cluster.

I. Rule Learning

Machine learning rules are also known as if else learning rules which is very simple to implement. This is very common in games because they need lot of repetitive and predictable tasks to complete. For example, if we have to complete the assignment it will take us some time and work, so it will be easier for us to complete it only if it is part of the assignment than we complete the assignment without it.

In the game, the computer will use the rules to help it to determine the most optimal plan for the player to complete the most tasks.

J. Evolutionary Algorithm

The Evolutionary algorithms is based on Genetic Algorithms. The main difference from Genetic algo algorithms is that instead of using parents to produce the children, Evolutionary Algorithms use the learning process to increase the value of parent’s genes. The main concept of the algorithm is that a lot of parents will breed an offspring which will be the one to produce the highest fitness in the offspring. The fitness of individuals that are produced is measured based on some criteria like speed, efficiency, accuracy etc.

K. Artificial Neural Networks

Artificial neural networks are very complex to implement, but they are extremely useful for a lot of problems. A good example for one can be Netflix Prize.