DEEP LEARNING FOR NATURAL LANGUAGE PROCESSING
Sem-II, 2024-25
Updates
- New Lecture is up: Responsible LLMs & Conclusion [slides]
- New Lecture is up: Knowledge Editing [slides]
- New Lecture is up: Tool Augmentation with LLMs [slides]
- New Lecture is up: Guest Lecture - Retrieval-Augmented Generation (Dinesh Raghu) [slides]
-
Join Piazza to get course notifications, access code: ell884dlnlp.
Course Description
Natural Language Processing (NLP) is the art of teaching a computer how to understand (human) language. It is a type of AI technology that allows an automated system to understand human needs through natural language. Anything that deals with human languages (texts or speeches), such as web search, recommendation systems, chatbots, language translation, and social media, comes under the realm of NLP. Due to the success of deep learning, in the last decade, NLP has witnessed a massive change in how a computer perceives a language. The blend of classical NLP with neural networks has shown promising performance in almost every application. This crash course intends to provide a holistic view of modern NLP techniques. It will start with two introductory sections -- basics of NLP and introduction to deep learning, following which it will broadly focus on deep learning methods for NLP. The course is designed for senior undergraduate and graduate students of any discipline, who carry introductory knowledge of machine learning and a strong Python programming skill. The students are not expected to carry any prior knowledge of NLP at the start of the course. We will cover foundational architectures, different training protocols, and several downstream NLP tasks. The course will include programming assignments (in Python), quizzes, course projects, research paper reading, and two written exams (midterm and major).
Prerequisites:
- Basic computer science principles (Big-O notation, Comfortably write non-trivial code in Python/numpy)
- Probability (Random Variables, Expectations, Distributions)
- Linear Algebra & Multivariate/Matrix Calculus (Gradients and Hessians, Eigenvalue/vector)
- Machine Learning
- Deep Learning
TA Availability
| Name | Projects | Time | Contact |
| Sahil | Project 7 | Tue@4-5 | sahil.mishra@ee.iitd.ac.in |
| Vaibhav | Projects 5 and 6 | Mon@4-5 | mt1210236@iitd.ac.in |
| Aswini | Projects 3 and 4 | Wed@3-4 | eez238359@ee.iitd.ac.in |
| Anwoy | Projects 1 and 2 | Thu@4-5 | eez238463@ee.iitd.ac.in |
Assessment Plan (Tentative)
- MidTerm: 20%
- Major: 30%
- 3x Quizzes: 15%
- 2x Assignment (to be done individually): 15%
- Project (group-wise): 18%
- Paper Reading (group-wise): 2%
Plagiarism Policy
- The assignments are supposed to be done individually. You should carry out all the implementation by yourself.
- The project is supposed to be done by members of a group only. Collaboration between groups is not allowed
- Any cheating will result in a zero on the assignment/project, an additional penalty of the negative of the total weightage of the assignment/project, and possibly much stricter penalties (including a failing grade and/or referring to a DisCo). Please note that using any AI tools is not allowed. If the software detects any plagiarism due to it, this will be penalized equivalently.
Tentative Penalties
- For greater than 30% Plagiarism, 0 in the assignment/project and -5% in the course total
- For GPT-generated code, penalties will be decided based on the extent of influence of the code and other factors li
- These are just tentative. Stricter penalties might be applied based on the degree of plagiarism
