google-research/uda

This project helps machine learning practitioners significantly reduce the amount of labeled data needed to train high-performing models for tasks like classifying images or text. You feed it a small set of labeled examples alongside a larger pool of unlabeled data, and it outputs a highly accurate classification model. Data scientists, machine learning engineers, and researchers can use this to build effective models even when manual data labeling is expensive or time-consuming.

2,202 stars. No commits in the last 6 months.