Overview
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Explore conditioned generation in neural networks for natural language processing through this comprehensive lecture from Carnegie Mellon University's 2018 Neural Nets for NLP course. Delve into key concepts including language models, conditioned language models, and various generation techniques such as ancestral sampling and ensembling. Learn about evaluation methods like human evaluation and perplexity, and gain insights into the differences between linear and log-linear models. Discover advanced topics such as parameter averaging, ensemble distillation, and stacking. Conclude with a contrastive look at evaluating unconditioned generation, providing a well-rounded understanding of this crucial aspect of NLP.
Syllabus
Intro
Language Models Language models are generative models of text
Conditioned Language Models
Ancestral Sampling
Ensembling
Linear or Log Linear?
Parameter Averaging
Ensemble Distillation (e.g. Kim et al. 2016)
Stacking
Basic Evaluation Paradigm
Human Evaluation
Perplexity
A Contrastive Note: Evaluating Unconditioned Generation
Taught by
Graham Neubig