• Go beyond being a LLM user and learn how to build and manipulate them from scratch.
• Build your own small and efficient custom models.

This course introduces in detail the machinery that makes Deep Learning work for NLP, including the latest transformer models and large language models like GPT. Attendants will be able to understand, modify and apply current and future Deep Learning models. They will learn the inner workings of the models, how to build smaller, more efficient models, and implement them in Keras.

Student profile: professionals, researchers and students with basic programming and Python experience. Basic math skills (algebra or pre-calculus) are also needed. Although not strictly necessary, we recommend subscribing to Collab Pro for more out of GPUs.

The course will introduce large language models, with special emphasis on adaptation techniques (e.g. in-context learning, few-shot, instruction learning) and ways to align with human preferences. In addition, advanced training techniques such as parallelism, selective architectures and scaling laws are presented.
Participants, in addition to understanding the fundamentals of LLMs and learning advanced training techniques, will gain hands-on experience in applying and working with these models, while addressing biases and ethical concerns.

Student profile: professionals, researchers and students with basic programming and Python experience. Basic math skills (algebra or pre-calculus) are also needed. Although not strictly necessary, we recommend subscribing to Collab Pro for more out of GPUs.

The aim of this course is to understand and deploy large language models (LLMs) from a practical perspective, enabling students to gain hands-on experience with these models without coding, using tools like Flowise and oLLaMa. Participants will learn how to use proprietary models and open-source models like Gemma or Qwen for prompt engineering, creating agents, chatbots, Retrieval-Augmented Generation (RAG) models, and other NLP applications. Additionally, non-generative tasks such as information retrieval will be covered. The course will also include multimodal models that incorporate images.
Participants will learn how to use proprietary models like GPT and open-source models like Qwen for prompt engineering, creating agents, chatbots, Retrieval Augmented Generation (RAG) models, and other NLP applications.

Student profile: This course is targeted at graduate students and professionals from various disciplines (linguistics, journalism, computer science, sociology, etc.) who need to understand and deploy LLMs easily. The goal is to provide participants with the autonomy to solve practical problems by understanding and deploying LLM-based applications in diverse and creative ways. No coding skills are required for the practical content, but basic installation skills and admin permissions are necessary.

This course introduces the main Deep Learning techniques used in state-of-the-art Speech Processing. Participants will learn the fundamental approaches behind key tasks such as automatic speech recognition, speaker recognition, language identification and speaker diarization.
The course will present the main neural network architectures used for speech, including convolutional, recurrent and transformer-based models, as well as common speech representations and training strategies. Through practical examples, attendees will learn how current systems are built and how to apply existing models and toolkits to real-world speech processing tasks.

Student profile: professionals, researchers and students with programming and Python experience. Math and signal processing knowledge (at the level of a BSc in Sciences or Engineering) is also recommended. Although not strictly necessary, we recommend subscribing to Collab Pro for more GPU availability.