Generative Artificial Intelligence for Advanced Automation
Public Concerné
Applied AI Skills from the first year of master, in particular:
- Basics in AI, ML and distributed learning.
- Basics in ML programming
Objectifs pédagogiques
The objectives of the module is to understand in detail Generative Artificial Intelligence technologies and how they can be served using networking technologies.
Capacité et compétences acquises
Applied AI Skills
Contenu de la formation
- Introduction to Generative AI fundamentals
- Generative AI Use-Cases and Applications
- GenAI Modeling and Scheduling
- Traffic Modeling and Network Accelleration protocols
Lectures These sessions are dedicated to acquiring the theoretical concepts and fundamentals of Generative AI.
- Introduction to Generative AI and foundational models (GANs, VAEs)
- What is Generative AI? (vs. Discriminative AI)
- Early statistical models (e.g., Markov Chains, HMMs)
- Generative Adversarial Networks (GANs): architecture, training, strengths, and weaknesses
- Variational Autoencoders (VAEs): architecture, training, and applications
- Transformer architectures and Large Language Models (LLMs)
- The Transformer architecture: self-attention, QKV vectors, and multi-head attention
- The Encoder-Decoder framework and its variants
- Large Language Models (LLMs): pre-training and fine-tuning paradigms
- Prompt engineering: principles and techniques
- Diffusion models and creative applications
- The forward and reverse diffusion processes
- Latent Diffusion Models (LDMs): concepts and efficiency
- Applications in image and video generation (e.g., text-to-image)
- Tools and frameworks for implementing diffusion models
- Ethical issues, challenges, and future trends
- Ethical implications: bias, misinformation (deepfakes), and intellectual property
- Current challenges: hallucinations, controllability, and computational cost
- Responsible AI principles
- Future trends: multimodal generation, efficiency, and scientific discovery
- GenAI traffic modeling and network accelleration
- Collective Communications
- Data-driven versus Traffic Model
- Transport protocols (RDMA, ROCE)
- Project assignment and beginning of practical work (1 lesson)
- Group project follow-up (2 lessons)
- Use-case seminars (1-3 lessons)
Description des modalités de validation
Lab/project reports, final exam.
En alternance ou en formation continue en journée :
Merci de consulter directement le secrétariat pédagogique de votre formation pour obtenir le planning.
En formation « à la carte Hors Temps de Travail », en présentiel, à distance ou hybride :
Voir les dates, les lieux d'enseignement et les modalités d'inscription sur les sites internet des centres régionaux qui proposent cette formation.
