Weak Supervision Learning Explained скачать в хорошем качестве

Weak Supervision Learning Explained

Getting labeled training data is almost always the hardest part of an NLP program. In this video I describe how to use label functions and weak supervision to programmatically generate labels for text classifiers. The idea is to automatically bootstrap labels based on key words or phrases used by experts familiar with the domain. This simple idea can be formalized into a powerful labeling framework (sometimes called data programming) that probabilistically assigns labels based on these pieces of “virtual evidence.” We’ve found this approach is especially powerful when paired with recent transformer-based NLP models. I’ll introduce these basic ideas, talk about the open-source library Snorkel that you can start using today, and also discuss some recent research in self-supervision that I’m really excited about. ------------------------------------------------------------------------------------------------------------------------------------------------ 📚 READING LIST: Read our book - Become an AI Company in 90 Days: https://www.prolego.com/become-an-ai-... Read our research report - AI Natives Among Us: https://www.prolego.com/ai-natives Read our comic book - Adventures in AI: shorturl.at/vHLUY Read our research report - Win the War for Data Science Talent: https://www.prolego.com/win-the-war-f... 📹 PROLEGO - GREATEST HITS: Prolego + MIT AI Bootcamp:    • MIT Bootcamps: How to identify business op...   High Level Overview of GPT-3:    • High-Level Tutorial of OpenAI's GPT-3 | GP...   GPT-2 Demo Produces the Next Games of Thrones book, 'The Winds of Winter':    • GPT-2 Demo Produces the Next Games of Thro...   Weak Supervision Learning Explained:    • Weak Supervision Learning Explained   Novelty with Transformers and Clustering:    • Novelty with Transformers and Clustering w...   Follow Prolego LinkedIn:   / prolego