Course Overview
- Location
- Howe 104
- Time
- Thursdays 3:30 - 6pm
- Instructor
- Zining Zhu
- Office Hours
- Thursday 1-3:30 PM at GN414 or by appointment
Nowadays the advance in the capabilities of NLP technologies lead to elevated concerns. Where do their capabilities emerge from the parameters and structures? How do they leverage the knowledge to make decisions? This seminar course introduces the most popular topics and tools that researchers use to understand the NLP system’s capabilities along multiple analysis levels. In this course, students will read, present and discuss papers from top conferences. Students will also gain hands-on experience in a project that explains an NLP system.
Syllabus
Course Objectives
This course familiarizes students with the emerging challenges and the advancements in Explainable NLP. By the end of this semester, students should be able to:- Identify at least three methods that explains a trained NLP model by stating its mechanisms.
- Implement one method that explains a trained NLP model by stating its mechanisms.
- Identify at least three desiderata for explanations.
- Implement one method that evaluates explanations following a desideratum.
Grading
- Participation: 50 in total. 5 points per week (50 maximum). To get the participation mark, a student submits a brief “summary and discussion point” for each of the papers that will be discussed in the week. The deadline is 24 hours before the lecture. The summary will be visible to the students presenting the papers.
- Project: 50 in total. 20 for proposal presentation, and 30 for final presentation.