Supervised, Unsupervised, and Reinforcement Learning

Published in

Machine Learning Community

3 min readAug 24, 2021

1. Supervised Learning:

Definition: In Supervised Learning, the Machine learns on the labeled dataset (ie; input labeled data and output labeled data) Where you explicitly can tell them this is the input data and this is how the output data must be. For eg: the model is trained with the input images of cats and dogs, for a new image, the model must predict either as a cat or a dog.
Problem: Mainly two types of problems: regression and classification. Classification is about predicting a categorical(discrete) variable whereas regression is about predicting a continuous quantity.
Dataset: Machine trains on a labeled dataset which means the algorithm already knows the type of output it will predict before training. For eg: classifying an iris dataset into three different species(Setosa, Versicolor, and Virginia)
Training Phase: The machine has labeled input and output data. The algorithm acts as a function to map input data to output data. After the algorithm is trained it is tested using the new data.
End Goal: Supervised Learning algorithm gives you the predicted outcome.
Algorithms: Linear Regression, Logistic Regression, Support Vector Machine, K Nearest Neighbour, Random Forest.
Applications: Speech Recognition, Spam Detection, Weather Forecasting, Object Recognition, etc.

Definition: In Unsupervised Learning, the Machine learns on the unlabelled data that means the machine has to learn without any supervision(learn on its own) and for that, it has to find the hidden pattern in the data to make predictions for the output.
Problem: Mostly used to solve association, clustering, and anomaly detection problems. Association problem is about finding patterns or co-occurrences in data. Given healthcare information, Clustering is used to cluster types of diseases based on the similarity of the information. Anomaly detection is used for tracking unusual activities.
Dataset: In Unsupervised Learning, the machine is only given the input data and is not directed anywhere. From the given information, the machine has to understand itself by learning patterns in the given data. For eg: given input images of humans and supercars, after clustering the output will form into two separate groups one containing humans and the other containing supercars.
Training Phase: Given the input, the machine has to figure out its pattern on its own.
End Goal: Unsupervised Learning is all about learning patterns and extracting useful insights.
Algorithms: K-Means, C-Means
Applications: Recommendation Systems(Netflix, Amazon, Youtube), Anomaly Detection, Fraud Detection, etc.

Definition: Reinforcement Learning depends on a learning agent. The agent interacts in an unknown environment by doing some actions and discovering some results as errors or rewards. And from this, the agent gets trained and ready to predict the new data given to it.
Problem: Here the input itself depends on the actions, the agent takes. In robotics, the robot is the agent and its surrounding is the environment so the input depends on the action it takes.
Dataset: In Reinforcement Learning, there is no predefined data, the input depends on the actions taken by the agent. The actions are recorded in the form of matrices and later serve as a memory to the agent.
Training Phase: As there is no predefined data, the whole reinforcement learning process is a training and testing phase. It has to learn everything on its own and it starts by exploring and collecting data.
End Goal: Reinforcement Learning is a clueless agent which figures out data itself by learning and exploring in an environment.
Algorithms: Q-Learning, SARSA
Applications: Self-driving cars, gaming, etc.