Imagine you have a dataset that contains 1M images, and a deep learning supervised task that requires a great amount of data to be annotated in order to train it.
The big question then becomes, do you send all 1M images for labelling? This would be time-consuming and not to mention very costly. What if your budget can cover only 10% of the data? How do you choose which items to send for annotating?

In this talk, I will describe my practical solution to this problem, as an Algorithm Developer at Dataloop, where I was handling dozens of datasets in various domains. Starting by using unsupervised methods in order to find “what matters” in a dataset, I will cover the differences between self-supervision and unsupervised learning, and demonstrate how the former succeeded where the latter failed. I will also show why you shouldn't be afraid of high-dimensional data and how you could train a neural network without a specific task.