Machine Learning is a very large field that aims to provide machines with human-like capabilities of learning and making predictions by acquiring skills or knowledge from historical data. Due to its rapidly expanding capabilities of handling even the most complex of tasks with little to no human intervention with the utmost robustness and speed, it has been a buzz in the software industry.
In summary, as quoted by Arthur Samuel, Machine Learning “gives computers the ability to learn without being explicitly programmed.” As depicted in the below figure, in stark contrast to traditional programming where data and program are fetched to the computer to get the desired output, a machine learning algorithm takes in data and output to the computer to get the desired programme.
Concept of Machine Learning
The process of converting experience into knowledge is called learning. It can be broadly classified into four categories, also called the types of machine learning, based on the type of the learning data and the algorithm.
Types of Learnings in Machine Learning
The different types of learnings in machine learning along with their use cases are defined below.
Learning #1: Supervised Learning: Supervised learning is probably the most commonly used type of learning used in machine learning. The learning data, also called the labelled data, contains the desired output and the objective is to find an algorithm that maps the inputs to the outputs to make accurate future predictions. Thus, in supervised learning, the algorithm trains on training data to produce a derived function that can be applied to predict a good label for never-before-seen testing data.
As depicted in the figure, known data and known label are passed to a computer which devises a model based on the input. When the new data is fetched to this model, it is supposed to accurately predict the output.
Supervised Machine Learning can further be divided into two types:
1. Supervised Learning Type #1: Classification
The training data contains defined discrete labels and the goal is to predict the discrete labels the testing data maps to. This classification can be binary or multi-class, based on the use case. For example, Mails are generally classified into spam, inbox, promotions, forums, updates, etc.
2. Supervised Learning Type #1: Regression
The training data has discrete continuous values instead of labels and the goal is to predict a value as close to the actual output according to the trend observed in the training data. A major focus lies on minimising the error value to as little as possible.
For example: Predicting real estate prices in a locality based on various features. The model can take into account a single feature or multiple features.
The following figure helps explain if a particular supervised learning problem is a classification problem or a regression problem.
Image Source: TowardsDataScience
Learning #2: Unsupervised Learning
In stark contrast to supervised learning, in unsupervised learning, the learning data, also called the unlabelled data, contains only the input parameters. The model learns the correlation between the different data points by itself. Thus, in supervised learning, the algorithm trains on unstructured and unlabelled training data to club similar data subsets together. These algorithms are extremely powerful in identifying patterns or trends in the data.
As depicted in the figure, raw data is passed to an unsupervised machine learning algorithm which divides the similar data in smaller subgroups.
Unsupervised learning can further be divided into two types:
- Unsupervised Learning Type #1: Clustering: The focus is on splitting the raw data into a pre-defined number of clusters that are different from other clusters but similar within the same cluster. For example, Recommender systems recommend items similar to the ones you have already shown a liking towards on the basis of various features.
- Unsupervised Learning Type #1: Association: The focus is on finding relations within the data set. For example: Finding out the relationship between the sale of one product with the sale of other products, like the association between milk and bread.
Learning #3: Semi-Supervised Learning
As the name suggests, semi-supervised learning is an amalgamation of both supervised and unsupervised machine learning techniques. The training data includes both labelled and unlabelled data, however the proportion of unlabelled data is generally higher as compared to its labelled data counterpart.
In semi-supervised learning, the unsupervised technique is used to divide the data into smaller sub groups, which is then fed to supervised techniques to generate accurate labelling.
For example: Classification of images datasets where only a small fraction of images have been labelled.
Learning #4: Reinforcement Learning
Reinforcement learning is probably the latest development in the field of machine learning where the model keeps improving its performance and accuracy incrementally with the help of a reward feedback, which can be positive or negative. As and when new data is fed into the system, the system learns and adds the data to its knowledge base, making it better trained and more experienced.
As depicted in the figure below, the agent and environment keep interacting with each other with additional reward feedback which helps improve the model incrementally. The agent observes the state, performs the action on the state and based on its performance, receives a reward from the environment on the basis of which the model is reinforced.
For example: Self driving cars’ models get better after each drive since they add the experience from each drive to its knowledge.
In conclusion, the different machine learning techniques help solve the different use cases brought forward by the problem statement and the underlying data. A grasp of all the different learnings and when to use one type of learning over another is useful to determine the kind of algorithm for your underlying model.
Frequently Asked Questions
The four most commonly used types of machine learning are:
1. Supervised Learning
2. Unsupervised Learning
3. Semi-Supervised Learning
4. Reinforcement Learning
Contrary to the popular belief, the learnings of Artificial Intelligence are quite different from the learnings in Machine Learning. These are as follows:
1. Reactive Machines: No memory, only responds to different stimuli
2. Limited Memory: Uses memory to learn and improve responses
3. Theory of Mind: Understand the needs of other intelligent entities
4. Self-Aware: Uses self-awareness to stimulate human-like intelligence
Machine learning is used in a variety of fields to stimulate human-like knowledge to solve redundant problems faster and with more accuracy. Some of the common applications of machine learning are:
1. Fraud detection
2. Recommendation systems
4. Personal assistants
5. Optical character recognition
By Saarthak Jain