Industry NLP: Multitask and Continual Learning - Adapters and Hypernetworks Approach
Data Science Conference via YouTube
Overview
Explore a 28-minute conference talk from Data Science Conference Europe 2023 that addresses the challenges companies face in implementing multiple NLP tasks while managing new client requirements and distribution shifts. Discover how to identify and implement suitable multi-task learning (MTL) methods that can be effectively applied to continual learning (CL) scenarios. Learn about the advantages and implementation of adapters and hypernetwork approaches, which offer superior alternatives to single-task learning in MTL-CL environments. Understand how these approaches enable faster model training, reduced storage requirements, seamless ML system integration, and continuous training while maintaining performance comparable to single-task learning. Delivered by Tin Ferkovic in Belgrade, this presentation provides practical insights for organizations looking to optimize their NLP implementations through advanced learning techniques.
Syllabus
Industry NLP: Multitask & Continual Learning | Tin Ferkovic | DSC Europe 23
Taught by
Data Science Conference