Supervised vs Unsupervised Learning Explained
Supervised learning:
Supervised learning uses labelled datasets.Based on this labelled data, system will predict the possible outcome.
We train the system by providing labelled data so that when system encounter real & unseen data, it applies the learning from labelled data basis the same, it will predict the possible outcome.
For example we will train the system by providing various images of cats to identify the properties of cat like how does a cat look like & height, length of a cat etc.After being trained, when any unseen image of cat is provided to system ,it will quickly identify that provided image is of a cat.
In this type of learning:
1) features are represented by X
2) labels are represented by Y
3) then system try to find out the relationship(f) between feature(X) and labels(Y)
Y = f(X)
One constraint of Supervised learning is that deep learning models need massive amount of data to train the system and having that much data labelled is also a challenge.Due to which Unsupervised learning comes into picture which we will explain in the next section.
Another constraint of supervised learning is that it works only for the data that is specific to a use case so it works on tasks on which model has been trained and model can’t be reused on other tasks.Again we will try to overcome this constraint with the help of unsupervised learning.
There are 2 types of supervised learning:
1) Classification
2) Regression
A) Classification is the process of creating a model that helps in separating the data into multiple categories.It is applicable for situations where we want to mark our data into some categories like identifying an email is a spam or not spam category or based on a health data of a person, predicting the category of that individual into sick or healthy OR identifying if it can be a rainy day or not based on weather data are few other examples.
B) Regression algorithms, model predicts the quantitative data like income estimations, height & weight estimations by age.So here we predict the numerical continuous data.
Unsupervised learning:
Unsupervised learning works on un-labelled data.Here system try to find the similarities in the datasets without labelling the data.
It mainly finds the common patterns in the datasets.Identifying these hidden patterns is the key task of unsupervised learning.
With this learning, system might come up with surprising correlation between data which we are not aware about since it uses a common pattern among all the input data supplied.
Since most of the real world data falls into unlabelled data, Unsupervised learning becomes more important since we do not need to have human involvement for labelling the data first before feeding it to the system.
Common algorithms are Clustering and Dimensionality Reduction.
1) Here features are represented by X
2) but since there is no labelled data so Y values are not available.
3) then system try to self discover the patterns which it uses to predict the possible outcomes.
Self supervised learning:
This is sub set of unsupervised learning since it takes un labelled data and converts it to labelled data to be used by other models.This is done by automating the process of generating labels so that explicit labelling is not required.
Check out for Generative AI.