Course manual 2023/2024

Objectives

The students can explain and motivate the fundamental principles and mechanisms behind Deep Learning’s past, present and future
The students can explain the major challenges, directions and active domains of research in the field of deep learning along with their advantages and disadvantages
The student can program, train and run deep learning models in a server environment.
The student can express research ideas described in publications in clear terms for readers unfamiliar with the details and devise promising follow-up research directions
The student is able to debug and critically assess deep learning methods from a practical-engineering, and mathematical-theoretic point of view.
The student can independently tackle new deep learning problems with well-reasoned combinations of existing approaches.

Teaching methods

Lecture
Laptop seminar
Self-study
Seminar

The main goal of the lecture is to introduce new materials and show the connection of the principles and methods with current research.

The goal of the practical sessions is to get the student working independently and become proficient in programming for deep learning.

Learning activities

Activity	Hours
Hoorcollege	28
Tentamen	2.5
Werkcollege	28
Self study	110
Total	168	(6 EC x 28 uur)

Attendance

This programme does not have requirements concerning attendance (OER part B).

Additional requirements for this course:

This programme does not have requirements concerning attendance (OER part B).

Assessment

Item and weight	Details
Final grade
1 (100%) Tentamen

The grading of this course will consist of two compulsory parts (code assignments and written exam) and one optional oral exam (viva voce).

The goal of this structure is to ideally assess the learning outcomes while providing an inclusive learning experience that caters to a variety of students' individual strengths.

The code assignments: The three code assignments will be graded using a rough scale of fail/pass/distinction. The final coding assignments grade will be calculated as the average of the three assignments. For calculation purposes these are requivalent to 5,8 and 10. The code assignments are were you get to apply what you have learned in the lectures in a practical setting and get familiar with the essential tools like PyTorch and working on a server environment.

The viva: The student can choose to partake in the viva until the 24nd November, and the viva will take place on the 29th. In the viva, the student will be asked about their submitted solutions to the coding assignments, as well as questions about the materials until the 25th November. In particular, the exam will be structured as follows: 0-5min: student presents how they have solved the problem, highlighting a part of their code that to them was the most the difficult. 5-7min: follow-up questions about the code provided. 7-10min: other questions pertaining to the content covered in the lectures and tutorials.

The written exam: This is a classical written exam that tests material of the whole course. Final grade calculation: If a student signs up to the viva voce, the grade will be calculated as: final grade = 2/6 (viva grade) + 1/6 (coding assignment grade) + 3/6 (written exam grade) If a student does not sign up to the viva: final grade = 1/4 (coding assignment grade) + 3/4 (written exam grade) Note that signing up to the viva automatically means that it will be used in the final grade calculation.

Students pass the course if the calculated grade is >=5.5+ , but the exam grade must be >=5.0. If it is below 5.0, the student may take the resit exam, which will replace the grade of the original exam. All deadlines are as outlined, late hand-ins are not accepted. In case of sickness for exams/assignments, please contact the student advisor.

Inspection of assessed work

Canvas' Ed will be kept open for discussions.

The ANS platform will be used for grading the students for assignments and exams.

Assignments

Assignment 1	Multilayer Perceptrons, Backpropagation,
Assignment 2	Pretrained models, ViTs, Prompt-learning
Assignment 3	Generative Models

Fraud and plagiarism

The 'Regulations governing fraud and plagiarism for UvA students' applies to this course. This will be monitored carefully. Upon suspicion of fraud or plagiarism the Examinations Board of the programme will be informed. For the 'Regulations governing fraud and plagiarism for UvA students' see: www.student.uva.nl

Course structure

Week	Topics	Material	Deadlines
1	Introduction to Deep Learning Feedforward Neural Networks	Textbooks: Deep Learning (Chapters 1, 6, 7-7.5) Understanding Deep Learning (Ch 1.1 - 1.2.1.2, 4, 7-7.3)
2	Deep Learning Optimisation	Textbooks: Deep Learning (Ch 7.8, 8) Understanding Deep Learning (Ch 6, 9-9.3.3)
3	Convolutional Neural Networks Modern Convolutional Networks	Textbooks: Deep Learning (Chapter 9) Understanding Deep Learning (Ch 10, 11)	17.11: Deadline Assignment 1
4	Transformers Graph Neural Networks	Textbooks: UDL book (Ch 12, 13) "Attention is all you need" by Vaswani et al. Dive into DL (Ch 11) "Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges" by Bronstein et al. (Ch 1, 2, 3, 5.1, 5.3)
5	Generative Modelling/GANs Guest Lecture: AI for Science (Max Welling)	Textbooks: Deep Learning (Chapter 19, 20)	28.11. Deadline Assignment 2
6	Guest Lecture: Deep Learning for 3D (Christian Rupprecht) Deep Variational Inference	Deep Learning (Chapter 19, 20) "Auto-Encoding Variational Bayes" by Kingma & Welling, UDL's Chapter 17	12.12. Deadline Assignment 3
7	Self-supervised and Multi-modal Learning	Papers: "Self-labelling via simultaneous clustering and representation learning" by Asano et al. "A Cookbook of Self-Supervised Learning" by Balestriero et al. Tbd
8	Exam	Exam on 18th December

Contact information

Coordinator

dr. Y.M. Asano

Owner	Master Artificial Intelligence
Coordinator	dr. Y.M. Asano
Part of	Master Artificial Intelligence, Master Logic,

Deep Learning 1