Most machine learning libraries and frameworks are designed with individual experiments in mind, but not the full machine learning pipeline. Machine learning engineers spend the majority of their time iterating through alternative designs and selecting features. However, the current state-of-the-art ignores this workflow and forces engineers to redo the entire training process every time a model is changed. Were building an incremental framework that allows researchers to maximally reuse old computations to speed up future iterations on design and feature selection.

Im excited about the SuperUROP program because it allows me to leverage my experience in machine learning and data science research to build new frameworks that make the whole pipeline more efficient.