CS 6012 Deep Learning: Foundations and Applications

Quick Links:
Topics
Tasks
All Scores
Evaluation
Announcements
Resources
Midsemester Course Eval 		Instructor: Sharat. Do NOT send email. Use piazza.
Office: Rekhi Bldg, A102
Office Hours: Tue 13:30 to 14:00
Lecture hours: W/F 11:00 AM (Slot 7)
Lab: There is no lab.
Venue: KR225
Teaching Assistants:
  • Recent Announcements (Last modified approximately at )
    1. Welcome to the course. Most links will not work yet.
    2. Our first class is on Wed, July 29 in KR225. Please be seated by 10:55 AM.
    3. Some registrations require instructor approval.
    4. All announcements will move to Piazza. Specific link to come.
  • Informal Course Overview: This is an institute elective on foundations and applications of deep learning, aimed at students outside the core AI/ML pipeline. It is best thought of as "deep learning for the rest of us": enough exposure to understand the ideas, vocabulary, opportunities, and risks of DL as it affects engineering and science, but not an in-depth specialist course. We will discuss a proper subset of the following topics, in no particular order.
    • Ethics and societal aspects: privacy, data governance, explainability and transparency, environmental and societal impact
    • Basics: function approximators, backpropagation, automatic differentiation, loss functions, fitting models, measuring performance, regularization, supervised, unsupervised, and reinforcement learning
    • Architectures: convolutional neural networks, residual networks, sequence models, attention, transformers, and graph neural networks
    • Generative AI: generative adversarial networks, flows, autoencoders, and diffusion models
    • Agentic AI, multimodal learning, large language models, retrieval-augmented learning, representation learning, and applications in computer vision, text systems, and scientific domains
  • Course Prerequisites (aka "who is allowed"):
    1. The informal prerequisite is that you should be a student, not a test-taker. (What's the difference?)
    2. Other than that, students are expected to have basic programming skills (programming with loops, pointers, structures, recursion), linear algebra and matrix methods.
    3. Audits are prohibited.
    4. There are no specific hard course prerequisites.
    5. The course is not available to students in the academic units than (CSE, CMInDS, EE, and IEOR).
    6. The course is open (modulo the above) to postgraduate students (again, in academic units other than (CSE, CMInDS, EE, and IEOR)).
    7. That said, the general goal is to "enable" committed students who wish to be part of this course. Please note that we are talking about students, and not test-takers. It's definitely not kosher to repeat this course with some other course that has large deep learning content. But modulo the above ("target audience") you are welcome.
  • What I will discuss next
    1. TBD
  • Tasks. Assignments are not optional. Piazza will have the latest on these tasks. (The following is a stub from previous offerings).
    1. Upload assignments to moodle. For group assignment, lowest (lexicographic) roll number submits inlab.

      In the situation that moodle is down around the deadline, make a copy, and upload it to Google drive or Dropbox, and send link. Do not send the assignment via email. All assignments will be posted on Piazza

    2. Task Zero:
    3. Open Project Project PartA
    4. Open Project Project PartB
  • Notes on evaluation.
    1. Grading. This is tentative. The final version will be available on Piazza. The following can be safely discarded after drop/add deadline.
      • Class participation (Viva): Pass/Not Pass (5%-10%) (starts after course drop/add).
      • Open Project.
      • Two pen-paper exam/quiz/ remaining percentage
      • Instructor Viva (not for everyone) to ensure you understand what you are doing. Viva can reduce marks for some people in group assignments.
      • Piazza Points: This will be used for generously tipping you when you are on the grade border.
    2. This is important . By default, I will assume that you will adhere to the following pledge (aka honour code)

      I pledge on my honour that I have not given or received any unauthorized assistance on this assignment or any previous task.

      and will write and sign this on EVERY submission that carries points. (You can use (Gita/Koran/Bible/Torah/Tanakh/Mother/BFF/<your choice>) (pick any one) instead of "honour" in the above line. If you are not clear what unauthorized assistance means, please talk to me.

    3. Collaboration: By default, i.e., unless specified otherwise, you may discuss general ideas behind tasks with people in the course. What does this mean?
      • In any case, you are not to discuss this with external people -- anyone other than TA -- outside your course.
      • Needless to say, you are not allowed to take previous year's question papers or assignments. This will also apply in the future after you finish with this course when your juniors ask you.
      • Any electronic discussion must happen via Piazza. No WA.
      • By default , please feel free to take things from the LLM/Internet but do not plagiarize from the Internet, i.e., if you do end up having to use external sources you MUST cite these clearly
      • For SOME tasks, you may be expressly prohibited to look for solutions from the Internet. In such cases, finding answers to problems on the Web or from other outside sources (these include anyone not enrolled in the class) is strictly forbidden. (Typically this is the general idea behind midsemester and final exams.)
      • Some tasks will be in groups. You can discuss within this group, i.e., for these tasks, the group is the unit
      • In all cases, you cannot borrow something for which you claim points. More specifically, if you submit something, and out of the 100% you submitted, 80% is from the Internet, that is fine -- only if you give proper acknowledgment. You will be entitled to be graded out of 100% even if your work of 20% is purely yours. But if you do not include acknowledgment for the 80% but you acknowledging only 30%, then you are falsely claiming credit
        The same situation if you are discussing your task with people on, say, WA.

      Violation of this policy is considered serious. By signing up for this course, and reading these lines, you agree to these terms :-)

    4. Even if you miss the task deadline, you have to submit your assignment to prevent getting a fail grade. (There is no guarantee that you will get any credit for it.)
    5. If you miss a submission deadline or an exam, your marks will be rescaled (based on other assignments) ONLY in exceptional circumstances (medical reason for example). These must be approved by me BEFORE the due date in writing or via email. The default is that you get zero.
    6. The following is true for all electronic submissions. Late submission carry a leaky exponential penalty. Assignments submitted within the first 29 minutes after the deadline are considered 'on time'. However any assignment that is late by more than 29 minutes will attract the penalty (table is as follows, intervals are semi-closed, i.e. open on the right). Thus an assignment submitted 31 minutes late w.r.t. the original, stated submission deadline will attract 20% penalty.
      Late submission in hours1-22-44-66-88-1010-1414-24late for more than 24 hours
      Penalty20%25%30%35%52%65%85%100% (You still have to submit it)
    7. Grade revision policy: Students have a habit for asking for more points on their exams. This is understandable. Please use the following policy for clarification of corrected papers:
      • If you have any questions on the grading, you must bring it to the attention of the professor concerned within 72 hours of receipt or the next lecture, whichever is earlier.
      • Please study the model answers before you question a decision. If you need to appeal a decision, please note that the instructor now has the ability to review prior marking, and thus continues to have the right to revise the marks of questions other than the ones you are debating.
      • I request that you do not ask frivolous questions. In particular, questions of the form "I think I should be given partial credit" are not welcome. You must instead use the objective criterion "Model answer states 2 marks for this step; I have written this step; please reevaluate".
  • Texts/References
    1. Simon J.D. Prince, Understanding Deep Learning, MIT Press, 2023.
    2. Antonio Torralba, Phillip Isola, William T. Freeman, Foundations of Computer Vision, MIT Press, 2024.
    3. Dan Hendrycks, Introduction to AI Safety, Ethics, and Society, 2024, Open Access (https://www.aisafetybook.com).
    4. Other References in no particular order
      • Ian Goodfellow, Yoshua Bengio, Aaron Courville, Deep Learning, MIT Press, 2016.
      • Christopher M. Bishop, Deep Learning: Foundations and Concepts, Springer, 2023.
      • David Foster, Generative Deep Learning, O'Reilly, 2022.
      • Sebastian Raschka, Build a Large Language Model (From Scratch), Manning, 2024.
      • Brian Christian, The Alignment Problem, W. W. Norton, 2020.
      • Rainer Muehlhoff, The Ethics of AI: Power, Critique, Responsibility, Bristol University Press, 2025, Open Access.
      • Bernd Carsten Stahl, Ethics of Artificial Intelligence: Case Studies and Options, Springer, 2021, Open Access.
      • UNESCO, Recommendation on the Ethics of Artificial Intelligence, UNESCO, 2021, Open Access.
  • Old Announcements
  • Solutions I used to post solutions, but nowadays I hand them out in class. Solutions may be occasionally posted and deleted asynchronously (in order that students from other courses do not suffer/benefit). 
  • Student Driven Course Evaluation.
    1. The anonymous midterm course evaluation process should happen week after midsem.