AID401 Deep Learning

Syllabus | International University of Sarajevo - Last Update on Oct 10, 2026

Referencing Curricula

HOSTED BY

Artificial Intelligence and Data Engineering

Academic Year

2026 - 2027

Semester

Fall

Course Code

AID401

Weekly Hours

3 Teaching + 2 Practice

ECTS

Prerequisites

CS404

Teaching Mode Delivery

Face-to-face

Prerequisite For

Teaching Mode Delivery Notes

Cycle

I Cycle

Ali Almisreb

Course Lecturer

Position

Associate Professor Dr.

aalmisreb@ius.edu.ba

Phone

033 957 243

Assistant(s)

Ismar Aganović

Assistant E-mail

iaganovic@ius.edu.ba

Course Objectives

Build a solid understanding of how deep neural networks are constructed and trained, including optimization methods, initialization, normalization, residual connections, and regularization techniques.
Develop practical skills in implementing, training, and debugging deep neural networks using PyTorch, as well as Keras and TensorFlow, on a variety of real-world datasets.
Introduce modern deep learning architectures and learning paradigms, including CNNs, RNNs, Transformers, GANs, diffusion models, self-supervised learning, and transfer learning, and analyze their strengths and limitations.
Equip students to apply deep learning to diverse tasks such as image classification, object detection, text generation, machine translation, speech recognition, and game playing, including fine-tuning pretrained and multimodal models.
Foster critical evaluation of deep learning systems with respect to performance, robustness, efficiency, and interpretability, and prepare students to deploy models responsibly in real-world applications.

Learning Outcomes

After successful completion of the course, the student will be able to:

1. Explain the main components and operations of a deep neural network, such as layers, activation functions, loss functions, optimizers, and regularization techniques.

2. Implement basic deep neural networks using Keras and TensorFlow and train them on various datasets.

3. Compare and contrast different types of deep neural networks, such as CNNs, RNNs, GANs, and transformers, and understand their advantages and disadvantages.

4. Apply deep neural networks to various tasks such as image classification, object detection, text generation, machine translation, speech recognition, and game playing.

5. Evaluate the performance and robustness of deep neural networks and troubleshoot common issues such as over

Course Materials

Required Textbook

Prince, S. J. D. (2023), Understanding Deep Learning, MIT Press. Available free online at udlbook.github.io/udlbook
Géron, A. (2025), Hands-On Machine Learning with Scikit-Learn and PyTorch, O'Reilly Media.

Additional Literature

Zhang, A., Lipton, Z. C., Li, M., & Smola, A. J. (2023), Dive into Deep Learning, Cambridge University Press. Available free online at d2l.ai
Raschka, S., Liu, Y., & Mirjalili, V. (2022), Machine Learning with PyTorch and Scikit-Learn, Packt Publishing.
PyTorch. PyTorch Tutorials. Official tutorials for object detection fine-tuning, DCGAN, deep reinforcement learning, and adversarial examples. pytorch.org/tutorials
Keras. Code Examples. Official tutorials for transfer learning and speech recognition. keras.io/examples

Teaching Methods

Lectures introduce deep learning concepts, architectures, and training techniques through visual explanations, worked derivations, and short live coding demonstrations

Class discussions compare methods and analyze model behavior, failure cases, and results

Supervised laboratory sessions provide guided implementation in PyTorch, Keras, and TensorFlow, with debugging support and continuous feedback

A semester-long final project develops the ability to formulate a problem, implement and evaluate a deep learning solution, and present the results

Weekly Topics

This weekly planning is subject to change with advance notice.

Week	Topic	Readings / References
1	Course overview; deep learning frameworks; building a tiny autograd engine.	T1 Chs. 1–4, 7; A1; A3
	Lab: Framework comparison: PyTorch autograd vs. TensorFlow GradientTape; the same model in PyTorch and Keras.
2 📍	Optimization in depth: SGD, momentum, Adam/AdamW, learning rate schedules, warmup, loss landscapes.	T1 Chs. 5–6; T2
	Lab: Optimizer and learning-rate comparison on Fashion-MNIST (PyTorch).
3	Making deep networks trainable: initialization, BatchNorm/LayerNorm/RMSNorm, residual connections, vanishing/exploding gradients.	T1 Chs. 7, 11; T2
	Lab: Deep MLP with and without normalization and residuals; observing vanishing gradients (PyTorch).
4	Generalization and training recipes: regularization, mixup/CutMix, label smoothing, double descent, debugging training.	T1 Chs. 8–9; T2
	Lab: Diagnosing and fixing overfitting; troubleshooting a broken training run (PyTorch).
5	Modern CNN architecture design: ResNet, depthwise separable convolutions, EfficientNet, ConvNeXt.	T1 Chs. 10–11; A1; A4
	Lab: Image classification with transfer learning (Keras Applications).
6	Attention from first principles: RNN/LSTM/GRU and the sequence bottleneck, the attention mechanism.	T2; A1; A2
	Lab: Character-level LSTM text generator (PyTorch).
7	Transformers in depth: self-attention, multi-head attention, positional encodings (incl. RoPE), KV cache.	T1 Ch. 12; T2; A1
	Lab: Transformer-based machine translation (PyTorch).
8	Midterm project submission/presentation	—
9	Self-supervised learning: contrastive learning (SimCLR), masked autoencoders, foundation models.	T1 Chs. 9, 14; A4
	Lab: Speech recognition: keyword spotting on Speech Commands (Keras/TensorFlow).
10	Transfer and adaptation: full fine-tuning, parameter-efficient fine-tuning (LoRA), instruction tuning overview.	T2; A3
	Lab: Fine-tuning a pretrained Faster R-CNN (torchvision); evaluation with IoU and mAP.
11	Diffusion models: DDPM, denoising intuition, latent diffusion, classifier-free guidance; comparison with VAEs and GANs.	T1 Chs. 15, 17–18; A3
	Lab: DCGAN implementation and comparison with diffusion samples (PyTorch).
12	Multimodal models: CLIP-style contrastive training, vision-language models.	T1 Ch. 19; T2; A3
	Lab: Deep reinforcement learning: DQN agent for game playing (PyTorch, Gymnasium).
13	Efficient deep learning and deployment: mixed precision, quantization, pruning, distillation, edge deployment.	T2; A3
	Lab: Model compression and edge deployment: PyTorch quantization and pruning vs. TensorFlow Lite conversion.
14	Trustworthy deep learning: adversarial examples, calibration, uncertainty, interpretability.	T1 Chs. 20–21; A3
	Lab: Adversarial attacks (FGSM), calibration analysis, Grad-CAM (PyTorch).
15	Final project presentations and course wrap-up.	—

Detailed Weekly Plan

Week 1: Course overview; deep learning frameworks; building a tiny autograd engine.

Lab: Framework comparison: PyTorch autograd vs. TensorFlow GradientTape; the same model in PyTorch and Keras.

Week 2: Optimization in depth: SGD, momentum, Adam/AdamW, learning rate schedules, warmup, loss landscapes.

Lab: Optimizer and learning-rate comparison on Fashion-MNIST (PyTorch).

Week 3: Making deep networks trainable: initialization, BatchNorm/LayerNorm/RMSNorm, residual connections, vanishing/exploding gradients.

Lab: Deep MLP with and without normalization and residuals; observing vanishing gradients (PyTorch).

Week 4: Generalization and training recipes: regularization, mixup/CutMix, label smoothing, double descent, debugging training.

Lab: Diagnosing and fixing overfitting; troubleshooting a broken training run (PyTorch).

Week 5: Modern CNN architecture design: ResNet, depthwise separable convolutions, EfficientNet, ConvNeXt.

Lab: Image classification with transfer learning (Keras Applications).

Week 6: Attention from first principles: RNN/LSTM/GRU and the sequence bottleneck, the attention mechanism.

Lab: Character-level LSTM text generator (PyTorch).

Week 7: Transformers in depth: self-attention, multi-head attention, positional encodings (incl. RoPE), KV cache.

Lab: Transformer-based machine translation (PyTorch).

Week 9: Self-supervised learning: contrastive learning (SimCLR), masked autoencoders, foundation models.

Lab: Speech recognition: keyword spotting on Speech Commands (Keras/TensorFlow).

Week 10: Transfer and adaptation: full fine-tuning, parameter-efficient fine-tuning (LoRA), instruction tuning overview.

Lab: Fine-tuning a pretrained Faster R-CNN (torchvision); evaluation with IoU and mAP.

Week 11: Diffusion models: DDPM, denoising intuition, latent diffusion, classifier-free guidance; comparison with VAEs and GANs.

Lab: DCGAN implementation and comparison with diffusion samples (PyTorch).

Week 12: Multimodal models: CLIP-style contrastive training, vision-language models.

Lab: Deep reinforcement learning: DQN agent for game playing (PyTorch, Gymnasium).

Week 13: Efficient deep learning and deployment: mixed precision, quantization, pruning, distillation, edge deployment.

Lab: Model compression and edge deployment: PyTorch quantization and pruning vs. TensorFlow Lite conversion.

Week 14: Trustworthy deep learning: adversarial examples, calibration, uncertainty, interpretability.

Lab: Adversarial attacks (FGSM), calibration analysis, Grad-CAM (PyTorch).

Course Schedule (All Sections)

Section	Type	Day 1	Venue 1	Day 2	Venue 2
AID401.1	Course	Friday 14:00 - 16:50	A F1.25	-	-
AID401.1	Tutorial	Friday 09:00 - 10:50	A F1.23	-	-

Office Hours & Room

Day	Time	Office	Notes
Monday	11:00 - 15:00	A F2.6
Thursday	11:00 - 15:00	A F2.6

Assessment Methods and Criteria

Assessment Components

40%x1

Final Exam

AI: Not Allowed

Alignment with Learning Outcomes : 1 2 3 4

30%x1

Midterm Project

AI: Allowed

Alignment with Learning Outcomes : 1 2

20%x1

Final Project

AI: Allowed

Alignment with Learning Outcomes : 3 4

10%x2

Assignments

AI: Allowed

Alignment with Learning Outcomes : 1 3

IUS Grading System

Grading Scale	IUS Grading System	IUS Coeff.	Letter (B&H)	Numerical (B&H)
0 - 44	F	0	F	5
45 - 54	E	1	F	5
55 - 64	C	2	E	6
65 - 69	C+	2.3	D	7
70 -74	B-	2.7	D	7
75 - 79	B	3	C	8
80 - 84	B+	3.3	C	8
85 - 94	A-	3.7	B	9
95 - 100	A	4	A	10

Late Work Policy

Information about late submission policies will be shared during class and posted in this section. Please check back for official guidelines.

ECTS Credit Calculation

📚 Student Workload

This 6 ECTS credit course corresponds to 150 hours of total student workload, distributed as follows:

Lectures

42 hours ⏳ (14 week × 3 h)

Laboratory sessions

26 hours ⏳ (13 week × 2 h)

Assignments

10 hours ⏳ (2 week × 5 h)

Midterm Project

20 hours ⏳ (1 week × 20 h)

Final Project (incl. presentation)

25 hours ⏳ (1 week × 25 h)

Final Exam preparation and exam

27 hours ⏳ (3 week × 9 h)

150 Total Workload Hours

6 ECTS Credits

Course Policies

Academic Integrity

All work submitted must be your own. Plagiarism, cheating, or any form of academic dishonesty will result in disciplinary action according to university policies. When in doubt about citation practices, consult the instructor.

Attendance Policy

Students are expected to adhere to the attendance requirements as outlined in the International University of Sarajevo Study Rules and Regulations. Excessive absences, whether excused or unexcused, may impact academic performance and eligibility for assessment. Mandatory sessions (e.g., labs, workshops) require attendance unless formally exempted. For detailed policies on absences, documentation, and penalties, please refer to the official university regulations.

Technology & AI Policy

Laptops/tablets may be used for note-taking only during lectures. Phones should be silenced and put away during all class sessions. Audio/video recording requires prior permission from the instructor.

Artificial Intelligence (AI) Usage: The use of AI tools (e.g., ChatGPT, Copilot, Gemini) varies by assessment component. Please refer to the AI usage indicator next to each assessment item in the Assessment Methods and Criteria section above. Submitting AI-generated content as your own work, where AI is not explicitly allowed, constitutes an academic integrity violation.

Communication Policy

All course-related communication should occur through official university channels (institutional email or SIS). Emails should include [AID401] in the subject line.

Academic Quality Assurance Policy

Course Academic Quality Assurance is achieved through Semester Student Survey. At the end of each academic year, the institution of higher education is obliged to evaluate work of the academic staff, or the success of realization of the curricula.

More info

Learning Tips

Engage Actively

Be prepared to contribute thoughtfully during class discussions, labs, or collaborative work. Active participation deepens understanding and encourages critical thinking.

Read and Review Purposefully

Complete assigned readings or prep materials before class. Take notes, highlight key ideas, and jot down questions. Aim to grasp core concepts and their applications—not just facts.

Think Critically in Assignments

Use course frameworks or methodologies to analyze problems, case studies, or projects. Begin early to allow time for reflection and refinement. Seek feedback to improve your work.

Ask Questions Early

Don’t hesitate to reach out when something is unclear. Use office hours, discussion boards, or peer networks to clarify concepts and stay on track.

Syllabus Last Updated on Oct 10, 2026 | International University of Sarajevo

Print Syllabus

Referencing Curricula Print this page

Course Code	Course Title			Weekly Hours*		ECTS		Weekly Class Schedule
				T	P
AID401	Deep Learning			3	2	6		Friday: 2:00-5:00
Prerequisite	CS404		It is a prerequisite to	-
Lecturer	Ali Almisreb		Office Hours / Room / Phone	Monday: 11:00-15:00 Thursday: 11:00-15:00 A F2.6 - 033 957 243
E-mail	aalmisreb@ius.edu.ba
Assistant	Ismar Aganović		Assistant E-mail	iaganovic@ius.edu.ba
Course Objectives	Build a solid understanding of how deep neural networks are constructed and trained, including optimization methods, initialization, normalization, residual connections, and regularization techniques. Develop practical skills in implementing, training, and debugging deep neural networks using PyTorch, as well as Keras and TensorFlow, on a variety of real-world datasets. Introduce modern deep learning architectures and learning paradigms, including CNNs, RNNs, Transformers, GANs, diffusion models, self-supervised learning, and transfer learning, and analyze their strengths and limitations. Equip students to apply deep learning to diverse tasks such as image classification, object detection, text generation, machine translation, speech recognition, and game playing, including fine-tuning pretrained and multimodal models. Foster critical evaluation of deep learning systems with respect to performance, robustness, efficiency, and interpretability, and prepare students to deploy models responsibly in real-world applications.
Textbook	Prince, S. J. D. (2023), Understanding Deep Learning, MIT Press. Available free online at udlbook.github.io/udlbook Géron, A. (2025), Hands-On Machine Learning with Scikit-Learn and PyTorch, O'Reilly Media.
Additional Literature	Zhang, A., Lipton, Z. C., Li, M., & Smola, A. J. (2023), Dive into Deep Learning, Cambridge University Press. Available free online at d2l.ai Raschka, S., Liu, Y., & Mirjalili, V. (2022), Machine Learning with PyTorch and Scikit-Learn, Packt Publishing. PyTorch. PyTorch Tutorials. Official tutorials for object detection fine-tuning, DCGAN, deep reinforcement learning, and adversarial examples. pytorch.org/tutorials Keras. Code Examples. Official tutorials for transfer learning and speech recognition. keras.io/examples
Learning Outcomes	After successful completion of the course, the student will be able to:
	1. Explain the main components and operations of a deep neural network, such as layers, activation functions, loss functions, optimizers, and regularization techniques. 2. Implement basic deep neural networks using Keras and TensorFlow and train them on various datasets. 3. Compare and contrast different types of deep neural networks, such as CNNs, RNNs, GANs, and transformers, and understand their advantages and disadvantages. 4. Apply deep neural networks to various tasks such as image classification, object detection, text generation, machine translation, speech recognition, and game playing. 5. Evaluate the performance and robustness of deep neural networks and troubleshoot common issues such as over
Teaching Methods	Lectures introduce deep learning concepts, architectures, and training techniques through visual explanations, worked derivations, and short live coding demonstrations. Class discussions compare methods and analyze model behavior, failure cases, and results. Supervised laboratory sessions provide guided implementation in PyTorch, Keras, and TensorFlow, with debugging support and continuous feedback. A semester-long final project develops the ability to formulate a problem, implement and evaluate a deep learning solution, and present the results.
Teaching Method Delivery	Face-to-face	Teaching Method Delivery Notes
WEEK	TOPIC						REFERENCE
Week 1	Course overview; deep learning frameworks; building a tiny autograd engine.						T1 Chs. 1–4, 7; A1; A3
Week 2	Optimization in depth: SGD, momentum, Adam/AdamW, learning rate schedules, warmup, loss landscapes.						T1 Chs. 5–6; T2
Week 3	Making deep networks trainable: initialization, BatchNorm/LayerNorm/RMSNorm, residual connections, vanishing/exploding gradients.						T1 Chs. 7, 11; T2
Week 4	Generalization and training recipes: regularization, mixup/CutMix, label smoothing, double descent, debugging training.						T1 Chs. 8–9; T2
Week 5	Modern CNN architecture design: ResNet, depthwise separable convolutions, EfficientNet, ConvNeXt.						T1 Chs. 10–11; A1; A4
Week 6	Attention from first principles: RNN/LSTM/GRU and the sequence bottleneck, the attention mechanism.						T2; A1; A2
Week 7	Transformers in depth: self-attention, multi-head attention, positional encodings (incl. RoPE), KV cache.						T1 Ch. 12; T2; A1
Week 8	Midterm project submission/presentation						—
Week 9	Self-supervised learning: contrastive learning (SimCLR), masked autoencoders, foundation models.						T1 Chs. 9, 14; A4
Week 10	Transfer and adaptation: full fine-tuning, parameter-efficient fine-tuning (LoRA), instruction tuning overview.						T2; A3
Week 11	Diffusion models: DDPM, denoising intuition, latent diffusion, classifier-free guidance; comparison with VAEs and GANs.						T1 Chs. 15, 17–18; A3
Week 12	Multimodal models: CLIP-style contrastive training, vision-language models.						T1 Ch. 19; T2; A3
Week 13	Efficient deep learning and deployment: mixed precision, quantization, pruning, distillation, edge deployment.						T2; A3
Week 14	Trustworthy deep learning: adversarial examples, calibration, uncertainty, interpretability.						T1 Chs. 20–21; A3
Week 15	Final project presentations and course wrap-up.						—

Assessment Methods and Criteria	Evaluation Tool	Quantity	Weight	Alignment with LOs	AI Usage
	Final Exam	1	40	1,2,3,4	Not Allowed
	Semester Evaluation Components
	Midterm Project	1	30	1,2	Allowed
	Final Project	1	20	3,4	Allowed
	Assignments	2	10	1,3	Allowed

* ECTS Credit Calculation *

Activity	Hours	Weeks	Student Workload Hours	Activity	Hours	Weeks	Student Workload Hours
Lectures	3	14	42	Laboratory sessions	2	13	26
Assignments	5	2	10	Midterm Project	20	1	20
Final Project (incl. presentation)	25	1	25	Final Exam preparation and exam	9	3	27
			0
				Total Workload Hours =			150
*T= Teaching, P= Practice				ECTS Credit =			6
Course Academic Quality Assurance: Semester Student Survey						Last Update Date: 05/10/2026

Detailed Weekly Plan
Week 1: Course overview; deep learning frameworks; building a tiny autograd engine.	Lab: Framework comparison: PyTorch autograd vs. TensorFlow GradientTape; the same model in PyTorch and Keras.
Week 2: Optimization in depth: SGD, momentum, Adam/AdamW, learning rate schedules, warmup, loss landscapes.	Lab: Optimizer and learning-rate comparison on Fashion-MNIST (PyTorch).
Week 3: Making deep networks trainable: initialization, BatchNorm/LayerNorm/RMSNorm, residual connections, vanishing/exploding gradients.	Lab: Deep MLP with and without normalization and residuals; observing vanishing gradients (PyTorch).
Week 4: Generalization and training recipes: regularization, mixup/CutMix, label smoothing, double descent, debugging training.	Lab: Diagnosing and fixing overfitting; troubleshooting a broken training run (PyTorch).
Week 5: Modern CNN architecture design: ResNet, depthwise separable convolutions, EfficientNet, ConvNeXt.	Lab: Image classification with transfer learning (Keras Applications).
Week 6: Attention from first principles: RNN/LSTM/GRU and the sequence bottleneck, the attention mechanism.	Lab: Character-level LSTM text generator (PyTorch).
Week 7: Transformers in depth: self-attention, multi-head attention, positional encodings (incl. RoPE), KV cache.	Lab: Transformer-based machine translation (PyTorch).
Week 9: Self-supervised learning: contrastive learning (SimCLR), masked autoencoders, foundation models.	Lab: Speech recognition: keyword spotting on Speech Commands (Keras/TensorFlow).
Week 10: Transfer and adaptation: full fine-tuning, parameter-efficient fine-tuning (LoRA), instruction tuning overview.	Lab: Fine-tuning a pretrained Faster R-CNN (torchvision); evaluation with IoU and mAP.
Week 11: Diffusion models: DDPM, denoising intuition, latent diffusion, classifier-free guidance; comparison with VAEs and GANs.	Lab: DCGAN implementation and comparison with diffusion samples (PyTorch).
Week 12: Multimodal models: CLIP-style contrastive training, vision-language models.	Lab: Deep reinforcement learning: DQN agent for game playing (PyTorch, Gymnasium).
Week 13: Efficient deep learning and deployment: mixed precision, quantization, pruning, distillation, edge deployment.	Lab: Model compression and edge deployment: PyTorch quantization and pruning vs. TensorFlow Lite conversion.
Week 14: Trustworthy deep learning: adversarial examples, calibration, uncertainty, interpretability.	Lab: Adversarial attacks (FGSM), calibration analysis, Grad-CAM (PyTorch).