Neural sequence-to-sequence models have emerged as a powerful tool to address many natural language processing tasks, such as translating text from one language to another, summarizing documents, generating answers to questions, or rewriting language in a different style. The main objective of the course is to gain an understanding of the state of the art techniques for designing, training, and using sequence-to-sequence models to generate natural language, with a focus on machine translation.
Students are expected to (1) read, present, and discuss a range of classic and recent papers drawn from the Natural Language Processing, Machine Learning and Linguistics literature, to understand the strengths and limitations of neural sequence-to-sequence models, and (2) put these ideas in practice in a semester-long research project.
While there are no formal prerequisites, this is an advanced topic course aimed at graduate students doing research in natural language processing, machine learning and related areas. We will assume that students are comfortable with the fundamentals of machine learning, deep neural networks, and natural language processing (e.g., CMSC723), or that they will catch up on their own. An early exam will help students assess their degree of preparation.
TR 3:30pm–4:45pm
IRB 1207
Contact: via Canvas for registered students
Office hours: Thursdays, 1:30-2:30pm, IRB 4130
Dennis Asamoah-Owusu
Contact: via Canvas for registered students
Assignments, grades, slides and all course commnications will be on Canvas