Bootleg: Guidable Self-Supervision for Named Entity Disambiguation

Explore a comprehensive lecture on Bootleg, a self-supervised system for Named Entity Disambiguation (NED), presented by Chris Re from Stanford University. Delve into the challenges of generalizing NED to rarely seen entities and learn how Bootleg improves tail performance using a transformer-based architecture. Discover the innovative inverse regularization scheme that enhances tail generalization and examine techniques for fixing systematic errors in self-supervised models. Gain insights into Bootleg's state-of-the-art performance on major NED benchmarks and its significant improvements over BERT baselines. Understand the real-world applications of Bootleg-like models and their impact on downstream applications. Presented at the Center for Language & Speech Processing (CLSP) at JHU, this 57-minute talk offers valuable knowledge for researchers and practitioners in the field of natural language processing and machine learning.