Understanding Self Supervised Learning, with Examples

Photo by Hadija on Unsplash

You must be familiar with the terms supervised and unsupervised learning. Somewhere between these two methods, we have something called self-supervised learning, which can solve really interesting problems.

To understand self-supervised learning properly, let’s briefly discuss supervised and unsupervised learning first.

Supervised learning

Let’s say, we have certain data(input) and labels(ground truth).

Now what we can do with this data is to train a model, be it a decision tree, neural network, to map the input to the labels. Labels act as a supervisor to the model.

Once the model is trained properly, it can predict the labels for the unlabelled data.

Classification and regression are the most common supervised learning tasks.

Classification | image by wallpaperaccess

Unsupervised learning

We have the data, but not the labels. Means no supervision.

But we can still perform a few tasks with unlabeled data. We can find similarities between data, segment the data, cluster the data into different…