• Networking: 127 h per year (Level L3) from 2013 to 2019.
• Mobile programming: 36 h per year (Level M2) from 2013 to 2019.
• Multimedia systems: 18 h per year (Level M2) from 2013 to 2019.
• Python and data science: 38 h per year (Level L2) from 2016 to 2019.
• System programming: 12 h per year (Level L3) shell, fork, shared memory, pthread, semaphore. From 2012 to 2014.
• Pattern recognition: 8h per year (Level M2) from 2015 to 2017.
• Industrial Networking: 12h per year. CAN protocol (Level M1) from 2015 to 2017.
• C for microcontroller: 10h per year (Level M1) from 2015 to 2017.

From Kantorovitch Problem to Linear Sum Assignment Problem

Graph Neural Network: a course and a notebook

• An illustration of Graph Neural Networks for graph classification
• A toy application to Letter classification
• A simple Graph Neural Network from Scratch (only with PyTorch and Networkx)

• Introduction to supervised Machine Learning: A probabilistic introduction. PDF slides
• Connecting local models : The case of chains PDF slides
• Graph Neural Networks. PDF slides
• The tutorial can be found here. HTML Notebook

A learning graph matching method based on integer linear programming

• An introduction to Learning Graph Matching.
• The graph matching problem is defined
• An integer linear program is presented
• A solver based on the integer linear program is implemented
• Integration of a combinatorial solver into a deep learning architecture.
• Backpropagation through a combinatorial solver is presented.
• An application to key point image matching is proposed.
• Code in Python base PyTorch Numpy, Matplotlib and Gurobi is proposed through a Jupyter Notebook.

No Graph Matching Solver

With Graph Matching Solver

---

A follow-up tutorial entitled "A Graph Neural Network from Scratch for semantic image segmentation".
Goals :

• An illustration of Graph Neural Network for node classification
• A toy application to skin detection
• A simple Graph Neural Network from Scratch (only with PyTorch and Networkx)
• The tutorial can be found here. HTML Notebook

A follow-up tutorial entitled "A Graph Neural Network (GNN) from Scratch for semi-supervised learning".
Goals :

• An illustration of GNN for semi-supervised learning
• A toy applicaion to community detection
• A lecture on Structured Machine Learning
• A simple Graph Neural Network from Scratch (only with PyTorch and Networkx)
• The tutorial can be found here. HTML Notebook

Supervised classification : Unsupervised classification : Semi-supervised classification :

---

Hidden Markov Model (HMM) for Named-Entity Recognition (NER)

• A presentation of HMM
• Formal definitions on inference and learning with HMM
• An application Named-Entity Recognition (NER)

• Introduction to supervised Machine Learning: A probabilistic introduction. PDF slides
• Connecting local models : The case of chains PDF slides
• The tutorial can be found here. HTML Notebook

Markov Random Field : A simple application to de-noising image

• An introduction to probabilistic graphical models.
• An application to Image Denoising.

• Introduction to supervised Machine Learning: A probabilistic introduction. PDF slides
• Connecting local models : The case of chains PDF slides
• Connecting local models : Beyond chains and trees.PDF slides
• The tutorial can be found here. HTML Notebook

Transfer Learning for object classification with an integration to an Android application

• Transfer learning: Fine tuning of MobileNet
• Integration an android application for custom object classification
• Speed comparison can be made according to computation device : CPU, GPU and NPU

A graph matching method based on the leading Eigen vector and Sinkhorn-Knopp algorithm

• An introduction to Graph Matching.
• The graph matching problem is defined
• A relaxed version of the graph matching problem is defined
• A solver based on the leading eigenvector is defined
• The obtained solution is refined by adding node-to-node constraints thanks to the Sinkhorn-Knopp algorithm
• An application to Letter matching is proposed.
• Code in Python base Numpy and Matplotlib is proposed through a Jupyter Notebook.

A graph matching method based on Integer Linear Programming

• An introduction to Graph Matching.
• The graph matching problem is defined
• An integer linear program is presented
• A solver based on the integer linear program is implemented
• An application to Letter matching is proposed.
• Code in Python base Numpy, Matplotlib and Gurobi is proposed through a Jupyter Notebook.

Reinforcement Learning by Q-Learning

• An introduction to Q-Learning.
• An application to train a robot to escape a maze.
• Code in Python base Numpy and Matplotlib.

• An introduction to Q-Learning.
• An application to sticks game.
• Code in Python base Numpy and Matplotlib.

• An introduction to Q-Learning.
• An application to Connect Four game.
• Code in Python base Numpy and Matplotlib.
• We explain the limitation of the Q-learning method.
• We draw conclusion on the solution based on Deep Reinforcement learning

The language of probability for machine learning: An application to skin detection thanks to the Naive bayes classifier

• An introduction to probability and the bayes' rule.
• An introduction to fitting probability models: Maximum likelihood, Maximum a posteriori.
• An introduction to generative and discriminative models.
• Introduction to supervised Machine Learning: A probabilistic introduction. PDF slides
• Connecting local models : The case of chains PDF slides
• Connecting local models : Beyond chains and trees.PDF slides

Simulation of self driving car

# Gradient Descent : Momentum, RMSProp and ADAM

Overfitting phenomenon of the maximum likelihood

Pedestrian image classification (in french)

Minimizing the cross entropy : a nice trip from Maximum likelihood to Kullback–Leibler divergence

• We show that minimizing the cross entropy is equivalent to maximizing the likelihood in case of discriminative models.
• We show that minimizing the cross entropy is equivalent to minimizing the Kullback–Leibler divergence
• We show that the cross entropy loss makes the logistic regression problem convex
• We present an application to cats and dogs image classification

Gaussian Mixture and Expectation Maximization algorithm

Minimizing the least squares error with quadratic regularization : but why

• we show that minimizing the least squares error in a regression context is related to a Gaussian assumption of the error distribution
• we also show that $L2$ regularization appears under an assumption of Gaussian distribution of the parameters and a Baysian treatment
• We show that maximizing the posterior distribution is equivalent to minimizing the regularized sum-of-squares error function with a quadratic regularizer

Maximizing the margin : Support Vector Machines

• Minimizing the cross entropy : a nice trip from Maximum likelihood to Kullback–Leibler divergence (http://romain.raveaux.free.fr/document/CrossEntropy.html)
• Overfitting phenomenon of the maximum likelihood (http://romain.raveaux.free.fr/document/Overfittingbiaisedandunbiaisedvariance.html)
• Gaussian Mixture and Expectation Maximization algorithm (http://romain.raveaux.free.fr/document/GaussianMixtureandExpectationMaximization.html)
• Relation between the least squares error the Maximum Likelihood (http://romain.raveaux.free.fr/document/LeastSquaresError.html)

• A data set is not well fitted by Gaussian models (under estimation of the variance)
• In Gaussian mixture models, a Gaussian distribution can be fitted to a single data point

My Ray Casting Tutorial

• This is a ray casting tutorial.
• The goal is to understand what is "ray casting" along with some piece of code (HTM5/Java Script).
• Latter, we want to use this technique to render a work site or a school.
• This tutorial serves as a basis for student projects.
Ray-casting is a pseudo 3-dimensional rendering technique that are very popular in game development arena in the 90s. Although ray-casting has been supplanted by newer and more powerful techniques (and hardware!), the reader can hopefully still benefit from the technique. Ray-casting sensation began with the release of a game, Wolfenstein 3D (iD Software), in 1992. In Wolfenstein 3D, the player is placed on a three dimensional maze-like environment, where he/she must find an exit while battling multiple opponents. Wolfenstein 3D becomes an instant classic for its fast and smooth animation. What enables this kind of animation is an innovative approach to three dimensional rendering known as “ray-casting.” Wolfenstein 3D was developed and created by Id Software. Henceforth, Id’s programmer, John Carmack, might well be the person who initiates the ray-casting sensation.

Ray-casting is a technique that transform a limited form of data (a very simplified map or floor plan) into a 3D projection by tracing rays from the view point into the viewing volume.





The tutorial can be found here. HTML

Helicoptere simulation

Graph in Pattern Recognition

Operating System Programing

Controller Area Network (CAN)

Android

Script Windows And Linux

Networking

Statistics

• Courses can be found here : Principal Component Analysis (PCA) and Correspondence Factor Analysis (CFA). PCA CFA
• Practical exercices I made can be downloaded here : TD1(PCA) and TD2(CFA).PCA exercices CFA exercices

Image and Ontology

• Here you can download the stand alone java application : Image2Owl. Image2Owl
• Here is the image ontology working with the software: ImageOntology.owl
• Here is a test image : shape_color.bmp
• Here is practical work about this : Tutorial

Computer Vision