Assignments and Projects Realized During the Master Mathématiques, Vision, Apprentissage (MVA) at ENS Paris-Saclay

First Semester

Convex Optimization (ALEXANDRE D'ASPREMONT)

• HM1: Convexity, Conjugate Function
• HM2: Duality
• HM3: LASSO

Computational Statistics (STÉPHANIE ALLASSONIÈRE)

• HM1: Markov Chains - Stochastic Gradient Descent
• HM2: Expectation-Maximisation Algorithm – Importance Sampling
• HM3: Hasting-Metropolis (and Gibbs) Samplers
• HM4: Metropolis-Hastings Algorithm

Image Denoising (JEAN-MICHEL MOREL, GABRIELE FACCIOLO, PABLO ARIAS)

• HM1: Ponomarenko Algorithm, Multi-Scale DCT
• HM2: BM3D, Non-Local Means
• HM3: Patch Similarity, Non-Local Bayes
• HM4: EPLL, Zoran-Weiss Gaussians Mixture Model
• HM5: Introduction to CNN Denoising
• HM6: A Deeper Understanding of CNN Denoising
• HM7: Noise to Noise
• HM8: Different Denoising Architectures

Medical Image Analysis (HERVÉ DELINGETTE, XAVIER PENNEC)

Paper Summary: Boundary Loss for Highly Unbalanced Segmentation
Reference: https://proceedings.mlr.press/v102/kervadec19a.html

Object Recognition and Computer Vision (IVAN LPATEV, JEAN PONCE, CORDELIA SCHMID, JOSEF SIVIC)

• HM1: Instance-Level Recognition
• HM2: Neural Networks
• HM3: Bird Image Classification Competition
  Kaggle Challenge: https://www.kaggle.com/competitions/mva-recvis-2021
• Project: Action-Conditioned 3D Human Motion Synthesis

Topological Data Analysis for Imaging and Machine Learning (FRÉDÉRIC CHAZAL, JULIEN TIERNY)

Second Semester

Deep Learning in Medical Imaging (OLIVIER COLLIOT, MARIA VAKALOPOULOU)

• Project: Prostate Cancer Diagnostic Using Histopathology Images
  Kaggle Challenge: https://www.kaggle.com/competitions/mvadlmi
• Paper Summary: Triplanar Ensemble U-Net Model for White Matter Hyperintensities Segmentation on MR Images
  Reference: https://www.sciencedirect.com/science/article/pii/S1361841521002309

Deep Learning in Practice (GUILLAUME CHARPIAT)

• HM1: Hyperparameters Tuning
• HM2: Grad-CAM
• HM3: Graph Neural Networks
• HM4: Small Data: Weak Supervision, Transfer, and Incorporation of Priors
• HM5: Koopman Decomposition, Duffing Oscillator
• HM6: Generative Models

Graphs in Machine Learning (DANIELE CALANDRIELLO)

• HM1: Spectral Clustering
• HM2: Semi-Supervised Learning (SSL)
• HM3: Introduction to Graph Neural Nets with JAX/Jraph

Time Series Analysis (LAURENT OUDRE)

• HM1: Convolutional Dictionary Learning (CDL), Dynamic Time Warping (DTW)
• HM2: ARIMA Process, Sparse Coding
• HM3: Change-Point Detection, Wavelet Transform for Graph Signals
• Project: Feature Selection Methods
  Reference: https://dl.acm.org/doi/pdf/10.1145/3136625