Deep Learning in Practice

• Most lectures: Guillaume Charpiat
• Practical sessions: Wenzhuo Liu and Nilo Schwencke (incl. materials by Victor Berger)

• January 5th : Deep learning vs. classical machine learning and optimization
  → Practical session: hyperparameters and training basics : instructions and the 4 exercises of TD1: Ex 1, Ex 2, Ex 3, Ex 4 with its dataset
  → Link to hand in TD1 (password communicated on the mailing-list)
  → Environment used in practical sessions
  → Practical tips and tricks to train neural networks
  → Course notes (pdf) (handwritten with drawings) and lesson summary (html)
  → Video recording: part 1 [90MB], part 2 [140MB], TD presentation [90MB]
  
  
• January 12th : Interpretability
  → Practical session: visualization with grad-CAM (including all side files) [73MB]
  → Link to hand in TD2 (same password as previously)
  → Course notes (pdf) (handwritten with drawings) and lesson summary (html)
  → Video recording: part 1 [200MB], part 2 [200MB], part 3 [230MB], TD presentation [170MB]
  
  
• January 19th : Architectures
  → Practical session on graph-NN: code and instructions; submission site (same password)
  → Course notes (pdf) (handwritten with drawings) and lesson summary
  → Video recording: full lecture [350MB] and TD presentation [115MB]
  
  
• January 26th : Small data: weak supervision, transfer, and incorporation of priors
  → Practical session on transfer learning: instructions and jupyter notebook; submission site
  → Video recording: full lecture [280MB] and TD4 presentation [100MB]
  → Course notes (pdf) (handwritten with drawings)
  → [2020] Lesson summary: part 1, part 2
  
  
• February 2nd, 9th, 16th : no course! but 2 sessions the following week:
  
  
• Monday February 22nd (9h-12h15 as well) : Guest talk + guest course
  → presentation of Therapixel, a start-up in medical imaging, by Yaroslav Nikulin
  → guest course by Olivier Teytaud (Facebook FAIR Paris) about Evolutionary Deep Computer Vision: slides + video recordings: part 1 [300MB], part 2 [500MB]
  
  
• February 23d : Modeling: deep learning and physics (exploiting known invariances, priors or physical properties), by Michele Alessandro Bucci (LRI, TAU team) and Lionel Mathelin (LIMSI, Paris-Sud)
  → Lionel Mathelin's slides and video recording [420MB]
  → Practical session on learning dynamical systems, with its dedicated submission site, and Alessandro Bucci's presentation [200MB]
  → Further on this topic: video recording [220MB] and associated notes
  → [2020] 2019's course on similar topics
  
  
• March 2nd : Generative models (GAN, VAE and Normalizing Flows) + Modeling tasks and losses
  → Practical session on generative models ; submission site
  → Video recordings: part 1 [170MB], part 2 [130MB], TP presentation + course complements [300MB], and associated notes (pdf)
  → [2020] Lesson summary about modeling tasks and losses
  
  
• March 9th : Guarantees? Generalization and formal proofs + Auto-ML / Auto-DeepLearning (guest talk by Lisheng Sun, from Isabelle Guyon's group)
  → Lesson about guarantees/robustness: video recording [230MB] and associated notes (pdf)
  → [2020] Lesson summary about learning from noisy data
  → Lisheng Sun's talk: AutoML slides and video recording [180MB] + more details about Bayesian optimization for hyper-parameters
  

• ours:
  • Deep Active Modeling for Physics
  • Capturing Molecular Conformations with Graph Neural Networks
  • Deep Learning for Fluid Mechanics
• more from the TAU team (deep for population genetics, AutoML, etc.)