Astron

A Weak Supervised Labeling System

This research project - Astron works on the problem of not having labeled data by utilizing the idea of labeling functions inspired by “Snorkel”,but focuses more on simplifying and standardizing the process. Astron requires human annotators to label a small portion of the data and will generate additional labeled data. Studies shows that Astron is able to label on average 15-20% of the data with 80+% accuracy using 5% of human labeled data.