Deep Learning with Small Data

I was very fortunate in first year to be able to undertake my own self-directed research project into any area that interested me. I picked machine learning, because I’d decided at this point that that was probably what I wanted to do with my life.

Since the field is really starting to expand out of the tech world and into other areas such as medical research, it’s difficult to obtain datasets big enough to train machine-learned models on them. This project loaded state of the art classifiers for MNIST and CIFAR-10, and looked at their performance when the amount of data provided to them was restricted in different ways.

I don’t think it’s my best work, but I think it was important for me to really get stuck into a project like this, which forced me to grapple with the various practicalities of machine learning and gain intuitions about its operation along the way.

You can read the paper here