Courses homeworks & projects

Course	Homeworks & Projects	Content
NLP (ENSAE)	Project and paper	Intent classification in Sequential labelling tasks, using contextual embeddings
Advanced Machine learning (ENSAE)	Final project	Understand and Fine-tune the ViT-Base/32 CLIP model
Data camp (IPP)	Data challenge	Solar wind classification based on data measured by in-situ spacecraft
	Design of Data challenge	Precipitation forecast: Based on 18 consecutive satellite radar frames, to predict the next 18 frames
Deep learning II (IPP)	Final project: RBM&DBN	Implement and train RBM (Restricted Boltzmann machine) and DBN (Deep belief network) from scratch
Altegrad (MVA)	Lab1	self-attention and HAN (Hierarchical Attention Network) architecture
	Lab2	Transfer learning on transformer architecture
	Lab3	Using Fairseq and HuggingFace transformers to finetune pretrained language models
	Lab4	Spectral Clustering for graphs; Graph Classification using Graph Kernels
	Lab5	DeepWalk algorithm & node embedding & Graph neural network (GNN)
	Lab6	Graph attention network (GAT) & Graph Classification with deep learning
	Lab7	DeepSets model & protein classification with GNN
	Kaggle challenge	Kaggle challenge: use sequential and structural information to classify protein into 18 classes.
Causal Inference (IPP)	Lectures & Labs	Notebooks in lectures and labs. See the summary in Readme
	Final project	Reproduction of paper: Counterfactual Fairness to study in machine learning the fairness using causal inference
Bootstrap (ENSAE)	TD1	Application of Jackknife to estimate the asymptotic variance (Ex.1) and bias (Ex.2) of estimators
	TD2	Application of Bootstrap to estimate the bias (Ex.2) and variance (Ex.1, possibly to use Boostrap of Boostrap) of an estimator/statistic
	TD3	The exam of last year
Sequential Monte-Carlo (ENSAE)	Final project	Employ the SMC methods in Dropout layer of neural network in adaptation stage, in order to replace the fine-tuning
Computer Vision (Telecom)	Lab1	self-attention and HAN (Hierarchical Attention Network) architecture
	Lab2	Feature detection (Harris corner detection) & Motion estimation (block matching) & Segmentation (algorithm of Otsu + region-growing based algorithm)
Data streaming (IPP)	Lab1	Discovery of River: like sklearn, but it focus on online machine learning
	Lab2	Using Docker and Kafka to analyse streaming tweets
	Final project: Continual GNN	Refactor the original codes in a paper studying streaming GNN via continual learning
Statistic Bayesian (ENSAE)	DM1	Application of MCMC Gibbs sampler to inference parameters based on 'a proteriori' probability
	Final project	Using Gibbs sampling and DMC-IS(direct monte carlo with importance sampling) to reproduce some results in this paper
Practical Machine learning (IPP)	Session 1	analyse on several unsupervised machine learning methods: K-means, GMM, PCA, t-SNE
	Session 2	reguralised regression, variable selection, nonlinear regression, on a dataset from the Brain Computer Interface competition
	Session 3	comparison between Bayesian decision, linear and nonlinear classification, on MNIST dataset and another one about diabetes
Deep learning (IPP)	Lab 1	implement MLP from scratch
	Lab 2	implement MLP using pytorch
	Lab 3	RNN (Many-to-one)
	Lab 4	a simple language model
	Lab 5	build CNN for image recognition, using Pytorch
	Lab 5	visualisation of CNN: Deep Dream algorithm; Adversarial examples
MAP566 Statistics in Action (X-3A)	Homework 1	Hypothesis testing
	Homework 1	implement MLP using pytorch
	TP3	Polynomial regression model
	TP4	Nonlinear regression model
	TP5	linear mixed model
	TP6	non linear mixed model
	TP7	mixture models
	TP8	Graph Clustering: Spectral and hierarchical methods
	TP9	Graph Clustering: Stochastic Blockmodels
MAP556 Monte Carlo Methods (X-3A)	TP1	Simulation of random variables + Law of large numbers + Central limit theorem
	TP2	Serveral methods of variance reduction: control variates, antithetic sampling, stratification
	TP3	Variance reduction through importance sampling
	TP5	Using Empirical Regression to approximate conditional expectation (in a context of finance)
	TP6	Generative Adversarial Network (GAN)
	TP8	Simulate processes of Brownian motion (eg. process of Ornstein-Uhlenbeck) and their Euler scheme
	TP9	Multi-level Monte-Carlo method (MLMC)
	Challenge1	simulate E(f(G)), f is reasonably regular
	Challenge2	play Angry Bird! Try to give a control on velocity to the bird facing a random wind
MAP553 Machine learning (X)	TP1	implement several optimization algorithms: GD, AGD, CGD, SGD, SAG, SVRG
	Project	a classic dataset of tree cover type classification, using auto machine learning, 2nd in Kaggle competition
Reinforcement learning (X-2A)	Lab3	Dynamic Programming - Value Iteration
	Lab4	Dynamic Programming - policy iteration
	Lab5	Temporal Difference
	Lab6	Q table - SARSA and Q Learning
	Lab7	Policy Gradient
INF580 Large scale mathematical optimization (X-3A)	Project	combine random projection and linear programming. Retrieve solutions from projected problem and dual projected problem, compute primal solution. Compare their feasibility error and compute time
MAP433 statistiques (X-2A)	TP1	Estimation parametrique. Loi de Poisson pour modéliser le nombre de buts marqués par une équipe de football
	TP2	Test de Cramer-von Mises
	TP3	Transformation de stabilisation de la variance
	Homework1	Estimation coefficients and interpretation (linear regression)
	Homework2	a test asymptotic on regression coefficients
	Homework3	Classification by KNN
MAP432 Markov & martingale (X-2A)	Project	L'algorithme du recuit simulé, pour résoudre des problèmes d'optimisation non convexe. On s'intéresse ici à une application au problème du voyageur de commerce
MAP435 optimisation (X-2A)	optimisation sans contrainte	Algorithme de gradient à pas fixe
		Algorithme du gradient pas optimal (le cas de fonction quadratic)
		Algorithme de Nesterov (fonctions convexes)
		Algorithme de Nesterov (fonctions fortement convexes)
		Algorithme du gradient conjugué
		Algorithme de Newton
		Analyse de vitesse de convergence et comparer les algos
		Quelques contre-exemples
	optimisation avec contraintes	Algorithme du gradient projeté
		Algorithme d'Uzawa
		Méthode de pénalisation
		Algorithme du Lagrangien augmenté