Overview
Learn about pretraining, finetuning, and uncertainty estimation in this comprehensive lecture from the University of Utah Data Science program, covering essential concepts like masked language modeling (MLM), language model finetuning for classification tasks, and pretrained vision models. Explore the fundamentals of uncertainty estimation, including probability calibration and calibration error metrics. Begin with MLM concepts and their applications, progress through the practical aspects of finetuning both language and vision models, and conclude with a thorough introduction to uncertainty estimation principles and their implementation in machine learning systems. Master the techniques for evaluating model confidence and reliability through calibration methods, while gaining insights into the broader implications of uncertainty quantification in AI applications.
Syllabus
Reminders
Masked language modeling MLM
Finetuning a MLM-pretrained model
Language modeling
Finetuning a LM for classification
Pretrained vision models
Uncertainty estimation intro
Probability calibration
Calibration error
Overview: uncertainty estimation so-far
Taught by
UofU Data Science