Reinforcement learning aims to teach an agent how to optimize its actions so as to maximize its gains.

This classic field has recently been revolutionized by Deep Learning (Q function, policy, etc.) and now makes it possible to solve tasks previously considered beyond the machine’s reach: process optimization, go/video games, robotics, etc.

Moyens techniques

Course material projected during training and sent to all trainees at the end of the course; case studies and practical examples chosen according to trainees' areas of interest

Suivi de l’exécution

All trainees are asked to sign in every half-day Evaluation: Questionnaire to assess skills acquired at the end of the course

Appréciation des résultats

Post-training satisfaction questionnaire

Objectifs pédagogiques

Theoretical courses mixed with examples and case studies. Participants are introduced to the key concepts of reinforcement learning and its recent developments.

Technologies abordées

Tensor, autograd, torch.n, Module, torch.optim, tensorboard, torchvision examples, TorchScript, torch.hub, torch.utils.dat

Compétences visées

  • Reinforcement learning: fundamentals
  • Recent variations: deep Q-learning
  • Model-free/model-based: case studies
  • Scaling up, state of the art