Deep Learning

General characterization

Prerequisites

Available soon

Bibliography

Zhang, A., Lipton, Z. C., Li, M., & Smola, A. J. (2023). Dive into deep learning. Cambridge University Press.   Bishop, C. M., & Bishop, H. (2023). Deep learning: Foundations and concepts. Springer Nature.   Bengio, Y., Goodfellow, I., & Courville, A. (2017). Deep learning. Cambridge, MA, USA: MIT press.

 

Teaching method

The Deep Learning course adopts an active, student-centered pedagogical model that integrates multimedia resources, collaborative platforms, and project-based learning to encourage deep understanding and practical skills.

 

Multimedia Resource Integration:

Each lecture provides resources on open platforms, including video lectures from international experts, blog posts by the instructor in the course webpage, and readings from books and research papers. This approach supports different learning styles and deep exploration of DL topics, promoting autonomy and diverse perspectives.

 

Interactive Lectures and Continuous Feedback:

Lectures are interactive, with the instructor frequently asking for student feedback and posing questions to encourage critical thinking. This active learning approach engages students, allowing them to reflect and participate in discussions. An anonymous feedback form is used to ensure the course content and delivery remain aligned with student needs.

Structured Online Collaboration:

A dedicated Discord server (or similar platform) is organized into channels covering lectures, tutorials, coding and math support, and office hours. Channels for team formation and assignment discussions promote collaboration and peer support. This setup fosters a community-based learning experience, enabling real-time Q&A, discussions, and teamwork beyond the classroom.

 

Project-Based and Hands-On Assignments:

The course emphasizes experiential learning through assignments based on real-world applications. Students engage in projects that require the implementation, adaptation, and optimization of deep learning models on diverse datasets (e.g., image, text, and graph data). These projects encourage the practical application of theory, reinforcing skills in data preprocessing, model building, tuning, and evaluation.

 

Active Support Through Tutorials and Office Hours:

Weekly practical sessions provide hands-on exercises, allowing students to apply learned concepts in coding and modelling contexts. These sessions are interactive, enabling students to solve problems collaboratively and receive immediate support from the instructor. Office hours are available for personalized support.

 

Peer Learning and Review:

Assignments include peer review, where students evaluate each other’s work based on the grading rubric, enhancing understanding through discussion, diverse perspectives, and critical analysis skills by assessing various approaches to similar problems.

 

Theory-Practice Integration and Ethical Reflection:

A core part of the pedagogical model is the integration of theory with practice. Concepts such as neural networks, CNNs, Transformers, and GNNs are taught through both lecture and practice sessions, ensuring that students not only understand the theory but also know how to implement and optimize the models. Additionally, discussions on trustworthy AI are present in the curriculum, encouraging students to reflect on the societal impact of DL.

Evaluation method

I. Assessment Components and Respective Weights

The evaluation consists of two components:

• Theoretical/Problems (T) – 50% of the final grade
• Project (P) – 50% of the final grade

II. Minimum Grades for Each Component

To pass, students must obtain:

• A minimum of 9.5/20 in the theoretical/problems component.
• A minimum of 9.5/20 in the project component.

III. Assessment Elements and Their Weights

Theoretical/Problems Component

• Two written tests, each accounting for 50% of the T grade.
• Alternatively, students may take a written exam consisting of two independent parts, each replacing the respective test grade if higher.
• Bonus: An extra point may be awarded based on submitting a report transcribing and deepening the content of one theoretical class OR by correctly responding to the questions of other students posted on the Discord discussion board on topics related to the course material.
• The grade formula for T is

T = round( min{20.0, 0.5 x Test1 + 0.5 x Test2 + ExtraPoint}, 1)

rounded to the first decimal.

Project Component

• Two assignments, each accounting for 50% of the P grade.
• The grade is individual and depends on the group work and individual reviews.
• The grade formula for P is

P = round(0.5 x Assignment1 + 0.5 x Assignment2, 1)

rounded to the first decimal.

IV. Minimum Grades for Assessment Elements

There are no minimum grades for individual tests, exam parts, or assignments. Check section II above for minimum grades for the T and P components.

V. Final Grade Calculation and Rounding Rules

• The final grade is computed as:

  Final Grade=0.5×T+0.5×P

• All individual component grades are rounded to one decimal place.

• The final grade is rounded to the nearest integer.

• If the final grade exceeds 17/20, students must take an oral exam to defend their grade. After the oral exam, the grade remains at least 17. The final grade is set to 17/20 if a student does not attend the grade defense.

VI. Conditions for Grade Defense

• Required for students with a computed final grade > 17.
• The grade remains at least 17 after the oral exam.
• The final grade is set to 17 if the student does not attend the grade defense.

VII. Conditions for Obtaining Frequency

Frequency is obtained by submitting the two assignments.

VIII. Frequency Validity

Frequency obtained since the 2021/2022 edition is valid for the current edition.

IX. Validity of Previous Classifications

The P grade obtained in past editions since 2021/2022 is valid for the current edition.

The instructor will query students to assess which students want to keep the past grade.

Past grades will not be kept if students register for tutorial classes.

X. Pre-registration for Assessment Elements

There is no pre-registration for tests and exams.

XI. Pre-registration Period

Not applicable.

XII. Admission Criteria for Students Without Pre-registration

Not applicable.

XIII. Permitted Assistive Devices

In written examinations, the following are not permitted:

• Calculators
• Reference materials
• Smartwatches
• Phones

In written examinations, the following are permitted:

• One A4 sheet of paper, handwritten by the student, possibly on both sides. In two-part examinations, two such sheets are allowed, one per part.
• Writing materials
• Simple watch
• ID card

 

Important: Rule approved by the Department of Computer Science Board in June 2025

During an assessment, a student may not have any electronic devices capable of accessing the internet or with Bluetooth connectivity (e.g., smartphones, smartwatches, smartglasses, tablets, laptops) with them, even if they are turned off.
Violation of this rule results in immediate failure of the curricular unit by exclusion and will be reported to the Scientific Committee of the respective program.

 

Subject matter

This course introduces key concepts and techniques in deep learning, moving from foundational principles to advanced architectures and applications.
-Fundamentals of Machine Learning
–Introduction
-Automatic differentiation and Multi-Layer Perceptron (MLP)
-Normalization, parameter initialization, and optimizers
-Convolutional Neural Networks (CNNs)
-Introduction to Computer vision
-Attention mechanisms, Transformers, and DL-based NLP models
-Introduction to Large Language Models
-Deep Reinforcement learning concepts and practical implementation
-Graph Neural Networks (GNNs) and uncertainty in predictions
-Diffusion models and trustworthy machine learning
Practical sessions provide hands-on experience, reinforcing theoretical concepts and developing skills in implementing and evaluating models.

Programs

Programs where the course is taught: