This is the page for the Deep Learning course (2027, תשפ”ז) at the SCE - Shamoon College of Engineering.
DRAFT - subject to changes.
| Week | Date | Mode | Topic | Material | Project |
|---|---|---|---|---|---|
| 1 | 22/10 | In-class | Course introduction, ML to DL, and project launch | Ch. 22, 23 | Groups formed; category proposals and vote |
| 2 | 29/10 | In-class | Vanilla neural network and how it trains | Ch. 21, 23 | Photography begins |
| 3 | 05/11 | Self-study (declaration) | Convolutional neural networks | Coursera C4W1 | Photography continues |
| 4 | 12/11 | Self-study (declaration) | Classic CNN architectures, transfer learning and augmentation | C4W2 | Upload form closes |
| 5 | 19/11 | Self-study (declaration) | Object detection | C4W3 | Dataset v1.0 on Kaggle; Part A starts |
| 6 | 26/11 | Self-study (declaration) | One-shot learning, Siamese networks and triplet loss | C4W4 | Part A |
| 7 | 03/12 | In-class | Augmentation, self-supervised representation learning and few-shot evaluation | Ch. 23 (augmentation), 26, 27 | Part A Part B starts |
| 8 | 10/12 | Self-study (declaration) | ML strategy: evaluation metrics, dev/test sets and human-level performance | C3W1 | Part A Part B |
| 9 | 17/12 | Self-study (declaration) | Error analysis, mislabeled data and data mismatch | C3W2 | Part A Part B |
| 10 | 24/12 | Project work (home/in-class) | Part A Part B | ||
| 11 | 31/12 | Project work (home/in-class) | Part A Part B | ||
| 12 | 07/01 | Project work (home/in-class) | Part A Part B | ||
| 13 | 14/01 | Presentations and defense (in-class) | Project ready |
“Ch.” refers to the chapters of the lecture notes; “C4W1” is week 1 of Coursera course C4, and so on.
In each self-study week, submit a short declaration on Moodle that you have studied the week’s material. The declaration carries no points. The self-study material is examined at the project defense, together with the material of the lectures.
Project description (Hebrew): HTML | PDF
The grade is based on the project only; there is no midterm or exam.
| Part | Who | Weight |
|---|---|---|
| Data | each student | 30% |
| Code | group | 50% (presentation and defense 35%, ranking between groups 15%) |
| Dataset error hunt | group, optional | 20%, protective |
Per student: final = 0.3·P + 0.5·D + 0.2·max(D, E), where P is the student’s data grade, D the group’s code grade and E the group’s error-hunt grade. The error hunt counts only if it raises the grade; a group that skips it has the code worth 70%.
The project defense includes questions on the lecture material.
LLM and agentic-AI tools are explicitly allowed for writing code and text. Understanding of the project and of the course material is tested at the defense.