SUBHANKAR MISHRA
ଶୁଭଙ୍କର ମିଶ୍ର
Reader-F, School of Computer Sciences, NISER
ପାଠକ-ଏଫ, ସଂଗଣକ ବିଜ୍ଞାନ ବିଦ୍ୟାଳୟ, ନାଇଜର
CS461/671 - Advanced Machine Learning 2023
Syllabus
- Regression, classification, regularization, gradient descent
- Neural Networks - Multilayer perceptron, Backpropagation, TensorFlow
- Images - Deep Learning - Convolutional Neural Networks, motivation, architecture
- NLP - Recurrent neural networks, backpropagation through time, long short term memory, attention
networks, memory networks
- Generative Models - Generative Adversarial Networks, Unsupervised learning, dimensionality reduction and
visualization.
Prerequisite
Grading scheme
Grading will be absolute. Total points = 100.
- Project : 40
- Assignment : 10
- Midterm [Theory + Programming]: 10
- Endterm [Theory + Programming]: 40
- Extra : 10 (Only when submitted to top conferences)
Lectures
| Week |
Topic |
Slide |
| Week 1 (Jul 31 - Aug 04) |
Course Logistics - Single Lecture only |
|
| Week 2 (Aug 07 - Aug 11) |
Motivation, background and usecases
CS Katha Barta Talk 11
|
|
| Week 3 (Aug 14 - Aug 18) |
Introduction to Deep Learning
CS Katha Barta Talk 12
|
MIT Slides
|
| Week 4 (Aug 21 - Aug 25) |
Walk through - implementation of Neural Networks
Code
CS Katha Barta Talk 13
|
|
| Week 5 (Aug 28 - Sep 01) |
Project - Proposal Presentation |
|
| Week 6 (Sep 04 - Sep 08) |
Deep Sequence Modeling ,
Code
CS Katha Barta Talk 14
|
MIT Slides |
| Week 7 (Sep 11 - Sep 15) |
Deep Computer Vision |
MIT Slides |
| Week 8 (Sep 18 - Sep 22) |
Midterm |
|
| Week 9 (Sep 25 - Sep 29) |
Holidays |
|
| Week 10, 11 (Oct 03 - Oct 13) |
Project - Midterm Presentation
CS Katha Barta Talk 15,
16, 17
|
|
| Week 12 (Oct 16 - Oct 20) |
Deep Generative Modeling,
Uncertainity and Bias
CS Katha Barta Talk 18
|
MIT Slides
MIT Slides
|
| Week 13 (Oct 23 - Oct 27) |
Deep Reinforcement Learning,
Limitations and New Frontiers |
MIT Slides
MIT Slides
|
| Week 14 (Oct 30 - Nov 03) |
Assignment Presentation |
|
| Week 15, 16 (Nov 06 - Nov 17) |
Project - Final Presentation |
|
Assignments
- [10] Assignment 1: Note
- Individual assignment
- [5] Cover the content of the material and present in the class along with Jupyter Demo
- [5] Illustrative Jupyter notebook for the algorithm (with explanation)
- Follow the github clone and pull request for both pdf, notebook and source .zip
Or
- [10] Shared Tasks
- Team of 2 people
- [10] If in Top 3 Or
- [5] If in Top 5 (Assuming participation is more than 10)
- Example - ACL Shared Tasks
Project
A group of maximum 2 persons. Class project must be something new that you did in this semester. Two types of
projects are
allowed:
- Category - 1: You pick/create an interesting (new) dataset, apply best suited one or more well known
machine
learning algorithms as baselines and extend these baselines in your creative and interesting way.
- Category - 2: A theoretical project should look at an open question, definite it concretely, look at
the literature, design and develop creative and interesting attempts.
Deliverables for reports will be in LaTeX generated pdf (with attached plagiarism report towards the end of the
report). Format should follow
ICML
template.
Updates:
- [05] (Max. 2 slides) Project Proposal - 5 mins
- Title, dataset, idea, relevant papers, teammate with work division, what to do by Midway, what
baselines to implement, what are the expected results
- [10] (Max. 10 slides) Project Midway - 15 mins + 5 mins Q&A, Midterm Report - Max Limit 3 pages
- [4] Insightful analysis of 2-3 related papers.
- [4] Experiments you have done along with results.
- [2] Report
- [25] Endterm Report - Max Limit 8 pages
- [15] New and interesting attempts to solve the problem
- [10] Report
- [10 Extra Points] Paper Submission by Dec 09, 2023, Top conference - Deadlines
| # |
Team Members |
Title |
Proposal |
Midterm |
Final |
Paper |
| 1 |
Sagar Prakash Barad |
Vision GNN-Powered Object Detection |
Slide |
Slide |
Slide |
Slide |
| 2 |
Aniket Nath |
Image generation using VICReg |
Slide |
Slide |
Slide |
Slide |
| Diptarko Choudhury |
| 3 |
Anna Binoy |
Predicting Flow Coefficients for Heavy Ion Collisions with Deep Learning |
Slide |
Slide |
Slide |
Slide |
| Arpan Maity |
| 4 |
Adhilsha A |
Oversampling in Heterogeneous Graphs using SMOTE |
Slide |
Slide |
Slide |
Slide |
| Deependra Singh |
| 5 |
Rahul Vishwakarma |
Using AI for Theorem Proving |
Slide |
Slide |
Slide |
Slide |
| 6 |
Aritra Mukhopadhyay |
Neural Networks at a Fraction: Table Structure Recognition |
Slide |
Slide |
Slide |
Slide |
Books
Some recommended books on Machine learning.
- The Elements of Statistical Learning Link
- MIT Introduction to Deep Learning
- Some of the
slides are burrowed from the course. Marking the copyright for the slides.
© Alexander Amini and Ava Amini
MIT Introduction to Deep Learning
IntroToDeepLearning.com
Academic Integrity
- For assignments you are allowed to discuss the assignments verbally with other class members, but you
are not allowed to look at or to copy anyone else's written solutions or code. All problem solutions and
code submitted must be material you have personally written during this quarter, except for any standard
library or utility functions.
- For class projects all reports submitted must be written by you or members of your project team. Code
generated for class projects can be a combination of code written by team members and publicly-available
code. You should clearly indicate in your reports and in your code documentation which parts of your
code was written by you or your team and which parts of your code was written by others.
- Academic honesty is a requirement for passing this class. Any student who compromises the academic
integrity of this course is subject to a failing grade. The work you submit must be your own. Academic
dishonesty includes, but is not limited to copying answers from another student, allowing another
student to copy your answers, communicating exam answers to other students during an exam, attempting to
use notes or other aids during an exam, or tampering with an exam after it has been corrected and then
returning it for more credit.