We may have heard about terms like Machine Learning and AI somewhere or other given their vast application in today’s world. It may seem as technical jargon to some and we may even use terms interchangeably. Machine Learning is a subfield of AI. Everything that comes under Machine Learning comes under AI. But everything that comes under AI does not come under Machine Learning.
Arthur Samuel who coined the term “Machine Learning” defined it as :“ Field of study that gives computers the ability to learn without being explicitly programmed”.
Tom Mitchell defined it as “ Computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E”. This defination can be better explained if we take an an example of playing chess.
E = the experience of playing many games of chess
T = the task of playing chess.
P = the probability that the program will win the next game.
Machine Learning builds a model based on data called as training data. This model is tested based on tested data. If after testing suitable accuracy is obtained then we can input new data whose prediction/classification is needed.
Machine Learning is classified into three categories:
• Supervised Learning
In Supervised learning the input data has labels along with other data features. For each data in the dataset along with the correct input there is correct output. So for each input we know what the output should look like.
These problems can be further divided into classification and regression.
In classification problems the outputs of all the inputs are in discrete form. The input can classified into either of the discrete categories. Let’s consider an example of “email filtering” where we have to classify an input mail either “spam“ or ”non-spam”. Here as we can see that the data can be classified into either to discrete two categories. Hence it is an example of classification.
In regression problems the outputs are in continuous form. There is no fixed no. of labels . Each input data may have different continuous output. Let’s consider an example of “house prices” where we have to predict price of a house given its features. Here there are no fixed no. of classed each input may have different output. Hence it is an example of regression
• Unsupervised Learning
In Unsupervised Learning there is no labels or output attached to each input. Machine can derive some sort of structure from data by clustering the data based on relationship between features. Unsupervised learning is not as popular as Supervised learning because of lack of accuracy and inconsistencies in unsupervised learning due lack of labeled output. Let’s consider an example where input dataset consists of features/characteristics of vegetables but no output/label . Unsupervised learning will automatically group these vegetables into groups that are somehow related based on their features.
• Reinforcement Learning
In reinforcement learning the model learns based on “reward” system. Whenever the model takes right step it is “rewarded”. This reward can be money ,winning. If model takes a wrong step then it is punished and the model will alter its behaviour. Through many steps of rewards and punishment the model will alter its behaviour to get more rewards and less punishments. More the number input, more efficient the model. Reinforcement learning has been used to make a machine play chess so well that they have defeated world champions!!
Top comments (0)