GA-Auto-PU: A genetic algorithm-based Auto-ML system for positive-unlabelled learning

What is it about?

GA-Auto-PU is the first automated machine learning (Auto-ML) system for positive-unlabelled (PU) learning. Many algorithms for PU learning have been proposed, so finding the optimal algorithm for a given task is computationally unfeasible. Auto-ML can assist in this task by searching for an optimal algorithm for a specific dataset through a guided process. We use a simple genetic algorithm to guide this process, and our system outperforms a state-of-the-art PU learning algorithm with statistical significance.

Featured Image

Photo by Kevin Ku on Unsplash

Photo by Kevin Ku on Unsplash

Why is it important?

PU learning is a field of machine learning that aims to learn classifiers from data consisting of a set of labelled positive instances, and a set of unlabelled instances, which may be positive or negative, but whose label is unknown. This is a common scenario in machine learning, but is often overlooked. In this case, often the unlabelled set is treated as a negative set, so instances are misrepresented. Our aim is to make PU learning more accessible, and therefore more widely used, by introducing an Auto-ML system that constructs PU learning algorithms for specific input datasets.