Overview
Explore advanced techniques for interpreting and debugging neural NLP models in this comprehensive lecture from CMU's Advanced NLP course. Delve into neural NLP model debugging methods, probing techniques, attribution methods, and interpretable evaluation. Learn how to identify and solve training time problems, optimization issues, and code-level challenges. Discover strategies for understanding model outputs, measuring accuracy, and interpreting predictions through saliency maps, gradient norms, and attention mechanisms. Gain valuable insights into improving model performance and enhancing interpretability in natural language processing tasks.
Syllabus
Introduction
Tools
Debugging
Training Time Problems
Optimization Problems
Mini Batch Loss Calculation
Structure Generation
Beam Size
Code Level
Early stopping
Understanding the outputs
Measuring accuracy
Interpretation of predictions
Saliency map
Gradient Norm
Attention
Taught by
Graham Neubig