What is Machine Learning

Machine learning Definition and History

Machine learning, a sub-discipline of computer science and a form of artificial intelligence (AI) that deals with the ability of the computers to learn and evolve without being programmed explicitly. In the year 1959, the term ‘machine learning’ was coined by Arthur lee Samuel (1901-1990). Arthur was an American computer expert and a pioneer in the field of AI and computer gaming. According to him, machine learning provides "computers the ability to learn without being explicitly programmed."

Machine learning is very similar to the cross-disciplinary subfield of computer science, popularly known as data mining. Though both the technologies look for patterns through data, they are significantly different. Data mining focuses on discovering the properties of data sets, whereas, machine learning is all about designing algorithms to learn from data automatically and make predictions. When it comes to data mining, the target is to extract data for human comprehension; but when it comes to machine learning, this data is used to detect patterns in data and adjust actions of the programs accordingly.

Machine learning is developed or evolved from pattern recognition and computational learning theory to artificial intelligence.

Machine learning is administered in a variety of computing tasks where programming and designing explicit algorithms with good performance is impractical or tricky. Typical use cases of machine learning may include detection of hackers or network intruders trying to exploit data breach, OCR (optical character recognition), email filtering and so on. Majority of the industries working with bulk amount of data have identified the value of machine learning technology. Organizations will be able to work more efficiently by collecting insights from this data in a shorter period.

Methods for Machine Learning

There are broadly four prominent methods for machine learning, which include: supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning.

Supervised Learning

Supervised Learning is the most popular method of implementing machine learning. When there are two different variables, the output variable (X) and the input variable (Y) and to learn the mapping function, an algorithm is used from the input to the output it can be defined as supervised learning. The target is to estimate the function of the mapping so as to receive new input data(Y) that can predict the output variables(X) of that particular data.

X=f(Y)

Here the algorithms are executed using labeled examples, in such a way that an input is provided where the desired output is known. Using methods like regression, classification, gradient boosting and prediction, this method uses patterns to foresee the values of the label on additional data’s that are unlabeled. This is used in applications that predict future events via the historical data, for example, to anticipate credit card transactions that are liable to be fraudulent or to predict the insurance customer who is most likely to file a claim etc.

Unsupervised Learning

Unlike the supervised learning, this method is used against data that doesn’t have historical labels. The system will be fed with the ‘right answer’; the algorithm has to figure out what is being shown. This method explores the data and finds the structure within. Popular techniques consist of nearest-neighbor mapping, self-organizing maps and singular value decomposition. Segment text topics, identify data outliers, and recommend items are also done via these algorithms.

Semi-supervised Learning

Semi-supervised learning is almost similar to supervised learning. It uses both unlabeled and labeled data for training. This method uses a major amount of unlabeled and minor amount of labeled data. Methods such as regression, prediction, and classification can be used for semi-supervised learning. This method is useful when the cost connected to labeling is increased for a fully labeled training process. Identifying an individual’s face on a webcam is an example of this method.

Reinforcement Learning 

Unlike the other three, reinforcement learning is used often for gaming, navigation, and robotics. Here the trial and error method is implemented by the algorithm to discover which actions yield the greatest reward. Learning of this type has three major components: the environment, the agent and the actions. The environment implies to anything the agent interacts with, the agent is the decision maker or the learner and action imply to what the agent is capable of doing. Through this method, the agent can prefer actions that in turn maximize the reward expected over a given period of time. By following a good policy the agent will reach the goal in a shorter period of time, thus the aim of the reinforcement learning is to learn the best policy available.