Solving Hard Problems When Data is Small - A Case Study with Semantic Parsing

Explore advanced techniques for tackling challenging natural language processing problems with limited labeled data in this 50-minute conference talk from the Toronto Machine Learning Series. Delve into cross-domain text-to-SQL semantic parsing for natural language database interfaces as Yanshuai Cao, Senior Research Lead at Borealis AI, shares insights on encoding prior knowledge in model architecture, training deep transformers on small datasets, and effective data augmentation strategies for NLP. Learn how to leverage task-specific unlabeled data and go beyond fine-tuning pre-trained models to bootstrap new systems when faced with scarce labels. Gain valuable knowledge on adapting machine learning approaches to scenarios where large-scale pre-training may not be sufficient, and discover techniques to enhance reasoning and quick adaptation capabilities in AI systems.