Course description

In this course, we will study contemporary machine learning methods for understanding and generation of human language. If you have taken machine learning and know how to implement, train, and deploy a classifier -- and now you want to understand how to bridge the gap between that and contemporary language models that can answer questions and hold a conversation, this course is for you. Through a series of homeworks, you will learn how to implement and train neural models to process language, from word-level embeddings to Transformer language models, including implementation, pre-training, fine-tuning and other modes of deployment of such models for handling different aspects of processing human language.

Pre-requisite: COMP 4220 / COMP 5220 Machine Learning or equivalent (with permission of instructor).

Class format

Each class will be divided into

• Lecture
• Practicum/Lab

There will be a 10-minute break after the lecture.

During the Practicum/Lab segment of the class, we will focus on technical (coding) skills and provide homework guidance. You will be expected to work on your homework during this time, so please bring your laptop.

Course materials

Class recordings will be available on Echo.

Class-related discussions and announcements will be conducted on Discord (see Blackboard for invite link).

COVID Safety

If you have any flu-like symptoms, or if you or somebody close to you have tested positive for COVID, please do not attend the class in person. If you are able to, please join the class meeting remotely. Please notify your instructor if you will need to join the class remotely.

Staff

	Name	Contact	Office	Office hours
Instructor	Anna Rumshisky	arumshisky@gmail.com	Dandeneau 318	TBA
TA	Namrata Shivagunde	namrata_shivagunde@student.uml.edu	Dandeneau 415	TBA
TA	Vijeta Deshpande	vijeta_deshpande@student.uml.edu	Dandeneau 415	TBA

Remote learning

Class recordings will be available on Echo (you need to log in with your University logon):

Echo recordings

Cheat sheets:

• Machine learning cheat sheet
• Review: Math preliminaries
• Python cheat sheet
• Numpy cheat shet
• Review: Mathematics for deep learning
• Github for beginners
• Github help

Grading

Homeworks	70%
Research Paper Presentations	10%
Midterm Oral Interview	5%
Final Oral Interview	15%

There will be no written final or midterm.

Homeworks

• We will have 6-7 homeworks.
• Homeworks are due at midnight on the day they are due, usually one day before the next class meeting.
• Homeworks will be posted on the course website and linked from the course schedule
• Homeworks must be submitted via Blackboard.

Research Paper Presentations

• Each student will be required to present a research paper assigned as readings for the class.

Late Policy:

• Homeworks will be accepted up to 2 (two) days after the original due date.
• Homeworks submitted up to 1 full day late will be graded at a 10% reduction.
• Homeworks submitted up to 2 full days late will be graded at a 20% reduction.
• After 2 days, Homeworks will not be accepted.

Collaboration Policy:

• Homeworks must be done individually.

Violating the collaboration policy by copying other people's work, as well as any other instance of cheating, including copying solutions from existing sources, carries the following penalties: (1) First violation leads to getting zero credit for the submitted assignment (2) Second violation leads to failing the course.