Introduction: Since 2012, Artificial Intelligence —reinvigorated by deep learning— has surpassed all previous methods in analyzing large-scale datasets (commonly known as big data) and solving complex tasks previously thought to require human intelligence. By 2022, these advances led to the release of ChatGPT, which became the fastest-growing app in history, reaching 100 million users in under one month — a milestone that took Google Search six years and Facebook four years to achieve. In 2024, tools like Stable Diffusion and MidJourney began generating images and videos that are nearly indistinguishable from real ones. The emergence of large language models and diffusion models has given rise to a new industry known as Generative AI (GenAI), now driven by major tech companies such as Google and Meta, as well as a vibrant ecosystem of startups led by OpenAI.
Beyond text, image, and video, AI is also transforming the natural sciences. For example, the 2024 Nobel Prize in Chemistry recognized breakthroughs in predicting the 3D structure of proteins — a long-standing challenge critical to understanding diseases and developing new drugs. AI tools like those from Google DeepMind and Anthropic are at the forefront of this work. Another promising direction is the design of new materials for applications in carbon capture, battery technology, and catalysis, contributing to solutions for global energy and environmental challenges.
Overview: Deep learning presents a fascinating paradox. On one hand, it can tackle highly complex tasks by extracting abstract and meaningful representations from data. On the other hand, the underlying algorithms are surprisingly accessible — both mathematically and computationally. In fact, the core mathematical concepts are simple, and implementing these techniques in Python is almost effortless thanks to modern libraries.
The goal of this course is to provide participants with a clear introduction, an intuitive understanding, and hands-on experience implementing the most successful deep learning techniques. The teaching approach strikes a careful balance between theory and practice. The theoretical component focuses on the fundamentals of deep neural networks, which are built upon simple linear algebra and basic gradient descent optimization. The practical component centers around PyTorch, the widely adopted deep learning framework developed by Meta AI.
Each lecture introduces key concepts and demonstrates how to translate them into working PyTorch code. The course focuses on the core classes of neural networks that underpin the majority of modern applications; Convolutional Neural Networks, Transformers (LLMs), and Diffusion Models. Together, these models cover approximately 95% of today’s deep learning use cases, spanning vision, language, and generative AI.
Objectives of the course
Why Take This Course? Embrace the Generative AI Revolution
• Understand the “why” behind the rise of Generative AI
• Learn the most influential architecture in AI today: the Transformer
• Gain a clear, intuitive, and solid understanding of neural network algorithms
• Implement generative models using Python notebooks
• Explore how leading tech companies apply deep learning in real-world industrial use cases
Target Audience:
This course is ideal for anyone who wants to:
• Get started with deep learning
• Apply deep learning to their own projects or research
• Learn how to code deep learning algorithms from scratch
• Upgrade their skills to the latest AI advancements, including Generative AI
Prerequisites
Participants should have:
• Basic knowledge of linear algebra (e.g. matrix multiplication)
• Familiarity with scripting languages such as Python, MATLAB, or R
The hands-on coding sessions will use Python and PyTorch.
Please bring your own laptop. No local installation is required, as all exercises will be run in cloud-based Python notebooks.
Course Materials
All course materials — including lecture slides and Python notebooks — will be made available one week before the course begins.
Copyright. All Rights Reserved.