Admission requirements
The course is open to students with interest in applying modern AI methods to the analysis of language. Some exposure to Python, computational linguistics and statistics is preferred, but not required.
Description
In this course, students will learn to apply deep learning methods to the analysis of natural language. We will discuss vectorized representations of texts (using a.o. word2vec and the recent BERT models), and we will apply neural networks to a variety of practical problems, including sentiment analysis, topic labeling, authorship attribution and question answering. The course consists of 12 lectures and practica. In the lectures, we will discuss the formal background of deep learning-based natural language processing, and recent relevant literature. In the practica, students will use an online environment (Google Colab) in which they can run their experiments. The practica begin with a short introduction to Python, limited to the fragment necessary to run the experiments (using the Keras deep learning library).
Course objectives
This course will teach students the fundamentals of deep learning methods for natural language processing (NLP), and will instill practical skills for implementing basis deep learning-based NLP systems. We will stay away from the intense mathematics of deep learning (although some of it will be discussed where necessary). After successful completion of the course, students are able to convert text into vectorized representations, and are able to apply deep learning-based algorithms to these representations. They will be able to understand from a functional perspective current literature on deep learning-based NLP.
Timetable
The timetables are available through My Timetable.
Mode of instruction
Lectures (12 * 2 hours; the first hour will be on background; the second hour on implementation).
Assessment method
Assessment
Oral group presentation and written exam with open questions.
Students will be evaluated through (1) a written exam (2) a joint presentation of a practical group assignment consisting of an implementation of an NLP topic (here, quality of solutions is not important; just a clear experimental narrative). Topics for the group assignment are voluntary, and will be chosen around 2/3 of the course, so that students can get timely assistance.
Weighing
The final grade will consist of an unweighted average of the scores for the written exam and the practicum.
Resit
The resit for the written exam will consist of an alternative partial written exam readdressing the topics of the failed questions. The resit for the group assignment will consist of a reprisal of the group presentation.
Inspection and feedback
How and when an exam review will take place will be disclosed together with the publication of the exam results at the latest. If a student requests a review within 30 days after publication of the exam results, an exam review will have to be organized.
Reading list
Stephan Raaijmakers: Deep learning for Natural Language Processing. Manning, 2020. Apply to teacher for eventually student discount.
Literature (papers), to be distributed at the onset of the course.
Python: http://www.spronck.net/pythonbook/index.xhtml
Registration
Enrolment through MyStudyMap is mandatory.
General information about course and exam enrolment is available on the website.
Contact
For substantive questions, contact the lecturer listed in the right information bar.
For questions about enrolment, admission, etc, contact the Education Administration Office: Reuvensplaats
Remarks
-