Georgia Tech – Fall 2020
Tuesday, Thursday 3:30 pm – 4:45 pm, Remote Synchronous Teaching
This is a TENTATIVE SYLLABUS, subject to change, as the pandemic situation develops during the semester and we know more about institute policies.
Instructor
Dr. Swati Gupta,
Assistant Professor and Fouts Family Early Career Professor,
School of Industrial and Systems Engineering,
Affiliations: Center for Machine Learning, ACO Program, Algorithms and Randomness Center, Physical Internet Lab, Parker H. Petit Institute, Institute for Data Engineering and Science
Office: Groseclose 437,
Email: swatig@gatech.edu
Office hours: TBD via survey for availability or schedule an appointment via email.
First class
The first class will meet synchronously to discuss the course content, policies and an introduction to online learning. The meeting link of the class is here: https://gatech.bluejeans.com/291572521
Remote Synchronous Teaching
All classes will be recorded and posted online on canvas. We plan to have synchronous lectures with quizzes during class time (see detailed plan below), and therefore, students are required to attend the online lectures. You can review the material using recordings, notes and reference books. We will also hold a survey in the first week of classes to determine the learning needs of all students: workload, time zone, being quarantined, and other situations that impact their learning. Adjustments may be made to the syllabus based on the findings of this survey.
Teaching Assistants
To be announced.
Recommended Texts
Notes for each class will be posted before the lecture. These notes are developed by Dr. Gupta and Jad Salem. This is a fairly new area of machine learning, and there are not many published books, but the following are useful references:
- Introduction to Online Convex Optimization by Elad Hazan
- Prediction, Learning, and Games by Nicolo Cesa-Bianchi and Gabor Lugosi.
Course Description
At the heart of most machine learning applications today – like advertisement placement, movie recommendation, and node prediction in evolving networks – is an optimization engine trying to provide the best decision with the information observed thus far in time, i.e. the problem of online learning. One must make online, real-time decisions and continuously improve the performance with the arrival of data and feedback from previous decisions. The course aims to provide a foundation for the development of such online methods and for their analysis. At times when these decisions are viewed in the context of the background of the consumers, there is some perceived bias or discrimination. The course will also discuss ethical and legal issues that arise due to such unintended consequences of decision-making and mathematical tools to formalize as well as mitigate such effects.
Outline
The course will have four interleaved components:
- fundamental theoretical tools for analyzing online methods, like regret analysis,
- algorithmic techniques for developing computationally efficient methods, like weighted majority algorithm,
- applications to real-world problems like computing the least congested path in a traffic network, and
- discussion around ethical and legal issues that might arise due to a perceived bias or discrimination in the decisions.
We will outline important connections to statistics, probability, optimization, graph theory, game theory and policy. We will also have some guest lectures by practitioners from the industry. Since the topic of the course is fairly novel, we will indicate directions of further research and exploration. This is an advanced class which involves development of the mathematics for why certain algorithms are useful and applicable as well as exposure to applying these algorithms to real-world datasets.
Learning Outcomes
At the end of this course, students will be able to do the following –
- Understand where online learning is applicable in many real-world scenarios,
- Develop algorithms that combine partial information as best as possible to make online decisions,
- Understand how exploration of decision space and exploitation from historic data must be prioritized to be able to reach optimal decisions,
- Understand the consequences of automated decisions on different population groups and how to mitigate harms caused by online machine learning.
Credits: 3-0-3.
Pre-requisites
The listed prerequisites are 3133 (Optimization) and 2027 (Probability). This is an advanced course, and there is a lot of subtle mathematics that we will cover to understand the basics of online machine learning. We will expect some mathematical maturity in linear algebra, set theory, and analysis as well. Assignment 0 covers the background needed for this class including basics of probability, set theory, analysis and optimization. If a student gets below 50% in Assignment 0, then special permission from the instructor is required to take the class. Further, assignments in this course are mathematical as well as programming-based and require a working knowledge of Jupyter notebooks and Python.
Grading policy
Due to uncertainty with the pandemic, we will allow students to drop their lowest two quiz grades.
| Assessment | Grade |
| Prerequisites: Assignment 0 | 3 |
| Assignment 1 | 10 |
| Assignment 2 | 10 |
| Assignment 3 | 10 |
| Quizzes (best 10 of 12) | 20 (2 points each) |
| Midterm I | 15 |
| Midterm II | 15 |
| 2 mandatary office hours with instructor or TA | 2 |
| Final Exam | 15 |
| Extra Credit | 2 |
| Total | 102 |
Course policies
- The assignments have to be submitted online – either with typed up solutions (word/LaTeX) or images of handwritten notes. Points will be taken off if the handwritten notes are not legible.
- For coding based questions, you have to submit the output as well as the code.
- You are encouraged to discuss homework/assignments problems with your fellow students. But your final answers should be based on your own understanding unless it is a group assignment, which will be announced on Canvas. Copying others’ work is NOT acceptable and violates the honor code. Write the names of students you have discussed the assignments with. In case answers are found copied from another student, both the students (the one who allowed copying, and the one who copied) will both get a 0 and might also result in obtaining a D or F grade overall.
- Regrade requests will be accepted via email to the instructor and the TA within 5 days of the graded assignment/exam being returned. Requests must include (1) a PDF of the original submission and (2) a LaTeX-produced PDF detailing which problems were graded incorrectly and an argument that the submitted solution is indeed correct. Regrades may only be requested if it is believed that a correct answer was marked as incorrect, not because insufficient partial credit was given to an incorrect or partially correct solution. If you request a regrade, you accept that the entire assignment/exam will be regraded, not just the problem(s) believed to be graded incorrectly.
- Late assignments: For every hour that the assignment is submitted late, the grade will be reduced by 10% (linearly interpolated), without exceptions. The only official extensions for late assignments will be granted with a letter from the Dean of Student’s Office to be received before the assignment is due.
Piazza/Canvas:
This course will be hosted on the Canvas. Homework assignments and solutions, and other announcements will be posted on Canvas. It is the responsibility of the students to keep notifications turned on for all announcements and updates. There is a Piazza for this course, which can be accessed through Canvas. Piazza is the best place to post any questions on the content or homeworks, so that other students or TA or the instructors can answer. Moreover, other students with similar questions will benefit from the discussion. You are strongly encouraged to take part in the discussions on Piazza, and high interaction will be rewarded using extra credit points.
Exams
Exams will be conducted virtually. Details of the exams will be announced later, and will depend on the virtual test-taking software available. Missing an exam will be accommodated only in case of Institute approved absences with a letter from the Dean of Students. Alternative arrangements will be made on a case by case basis, but the instructor must be informed at least two weeks prior to the exam.
Academic Honor Code
It is your responsibility to get familiar with the Georgia Tech Honor Code and you are bound by its requirements. For any questions any Academic Honor Code issues, please consult me, my teaching assistants, or www.honor.gatech.edu.
List of Topics and Class Plan
Despite the uncertainty due to pandemic, we will try to follow the class plan as closely as possible. However, view this as a tentative plan, subject to change, as the semester progresses.
| Week | Topic | Assignments and Quizzes |
| August 18, 20 | Warm-up and Introduction to Online Learning – Full information, partial feedback and ethics in decision making. | Assignment 0 released. Due on 21st August. In-class quiz 1 on August 20. |
| Module I – Full Information Setting | ||
| August 25, 27 | Learning with perfect and imperfect experts, Randomized v/s Deterministic | In-class quiz 2 on August 27. |
| September 1, 3 | Multiplicative Weights Update (MWU) | Assignment 1 released (due in two weeks). In-class quiz 3 on September 3rd. . |
| September 8, 10 | Solving linear programs using MWU | In-class quiz 4 on September 10th |
| September 15, 17 | Two-player Zero-sum Games and Review | In-class quiz 5 on September 17th |
| September 22 | Midterm I | All full information topics |
| Module II – Partial Feedback Setting | ||
| September 24, 29, October 1 | Exploration v/s Exploitation, Multi-armed Bandits Algorithms | Assignment 2 released (due in two weeks). In-class quiz 6 on October 1st. |
| October 6, 8 | Stochastic Bandits using Thompson Sampling and Upper Confidence Bounds | In-class quiz 7 on October 8th. |
| October 13, 15 | Stochastic Gradient Descent | In-class quiz 8 on October 15th. |
| October 20, 22 | Special topics: Deep Neural Networks or Recommendation Systems or Online Portfolio Optimization (TBD) | In-class quiz 9 on October 22nd. |
| October 27 | Midterm II (partial feedback topics except special topics) | |
| Module III – Ethics of Automated Decisions | ||
| October 29, November 3, 5 | Fair, biased or optimized decisions, Trade-offs in Metrics | Assignment 3 released (due in two weeks). In-class quiz 10 on November 5th. |
| November 10, 12 | Case Studies: COMPAS Risk Scores and Online Hiring | In-class quiz 11 on November 12th, |
| November 17, 19 | Guest Lectures from Industry Practitioners | No quiz this week. |
| November 24 | Review of the class | In-class quiz 12 on November 24th. |
| Final Exam | All topics, except special topics and guest lectures. Format to be decided. |
Mutual Respect and Expectations
ISyE 4803 Online Learning is an upper level theoretical course. The teaching team will try to provide best possible learning environment for you. Past students have commented the usefulness of the contents on their jobs, graduate school applications or senior design projects. You are expected to have an attitude to learn, instead of just getting a grade.
To produce a positive teaching and learning environment, instructors and students must partner with one another in and out of the classroom. Mutual respect is at the heart of such a partnership and is characterized by respectful language and imagery, punctuality and care for others’ time, clear and thorough communication, access to resources, and an openness to dialogue and debate. As a Georgia Tech faculty member, I am committed to such respect and I invite you to join me in working towards the best possible learning environment, so that all can meet their highest ambitions. Further, we will not tolerate discrimination towards any group of students. For more information about faculty and student expectations, as recognized in Georgia Tech policy, please see the following web page: http://www.catalog.gatech.edu/rules/22.php
Office of Disability Services
If you are a student with learning needs that require special accommodation, please contact the Office of Disability Services (http://disabilityservices.gatech.edu/) at (404)894-2563 as soon as possible to make an appointment to discuss your special needs and to obtain an accommodations letter. Please also e-mail me as soon as possible in order to set up a time to discuss your learning needs.
Comments are closed.