Acquis d'apprentissage visés
- Identify and select the appropriate machine learning algorithm family (supervised, unsupervised, deep learning) for a given problem, justifying this choice using objective criteria
- Implement, train, and evaluate classification and regression models by applying best practices: train/test/validation split, cross-validation, hyperparameter tuning
- Analyze and interpret model performance using appropriate metrics (precision, recall, F1, AUROC, RMSE, ) and diagnose bias and variance issues
- Apply unsupervised learning algorithms for dimensionality reduction (PCA, ICA, NMF) and clustering (k-means, GMM, hierarchical clustering) on real-world datasets
- Apply multimedia data analysis methods (images, signals) using suitable machine learning models
- Design, develop, and integrate tools and applications leveraging machine learning models for data valorization
Prérequis
- Probability and Random Variables (S5): probability distributions, expectation, variance, law of large numbers — foundations of probabilistic models (Naive Bayes, GMM, regression).
- Stochastic Processes (S6): modeling of time series and random signals — useful for sequential models and deep learning.
- Data Mining (S6): data preprocessing, association rules, introduction to clustering — direct foundation for this course.
- Expected cross-disciplinary skills:
- Intermediate-level Python programming (numpy, pandas, matplotlib)
- Linear algebra basics (vectors, matrices, eigenvalues) and differential calculus (gradient)
- Understanding of data structures and basic algorithms
- Ability to read technical documentation and scientific articles in English
Programme
- Foundations of Machine Learning:
- Overview of paradigms: supervised, unsupervised, reinforcement learning, and deep learning.
- Symbolic vs. numerical learning; statistical vs. structural/syntactic approaches.
- Learning as an optimization problem; data exploration through machine learning.
- Cross-cutting challenges: data quality, regularization, generalization, curse of dimensionality.
- Supervised Learning:
- Regression and classification tasks: use cases and selection criteria.
- Bias/variance trade-off; model complexity and generalization ability.
- Evaluation protocols: train/test/validation split, cross-validation, hyperparameter tuning.
- Classification metrics: precision, recall, F1, AUROC, confusion matrix; regression metrics: RMSE, MAE, .
- Fundamental algorithms: linear and logistic regression, KNN, Naive Bayes, decision trees.
- Extensions and ensemble methods: bagging, boosting, random forests, polynomial features.
- Model diagnosis: high bias vs. high variance; feature selection and engineering strategies.
- Multiclass classification; advanced algorithm: SVM (Support Vector Machine).
- Unsupervised Learning:
- Main tasks: clustering and dimensionality reduction.
- Clustering algorithms: k-means, hierarchical clustering (connectivity vs. centroid), DBSCAN, Gaussian Mixture Models (GMM).
- Dimensionality reduction algorithms: PCA (Principal Component Analysis), ICA (Independent Component Analysis), NMF (Non-negative Matrix Factorization).
- Evaluating unsupervised learning results: silhouette scores, inertia, interpretation of components.
- Deep Learning — Introduction:
- Artificial neural networks: architecture, activation functions, backpropagation.
- Convolutional networks (CNN) for image analysis; recurrent networks (RNN/LSTM) for signals and time series.
- Tools and frameworks: TensorFlow, PyTorch, Keras.
- Practical challenges and implementation:
- Data preprocessing: normalization, handling missing values, encoding categorical variables.
- Machine learning pipelines with scikit-learn.
- Model interpretability: SHAP, feature importance.
- Model deployment: prediction APIs, MLOps best practices.
Modalités d'évaluation
Continuous assessments and evaluated practical work.
Bibliographie
- Barra, V., Miclet, L. and Cornuéjols, A. - Artificial Learning — Concepts and Algorithms, from Bayes and Hume to Deep Learning - Eyrolles, 4th ed.
- Aurélien Géron - Hands-On Machine Learning with Scikit-Learn, Keras and TensorFlow - O’Reilly, 3rd ed., 2022
- Christopher M. Bishop - Pattern Recognition and Machine Learning - Springer, 2006
- Ian Goodfellow, Yoshua Bengio & Aaron Courville - Deep Learning - MIT Press, 2016: <https://www.deeplearningbook.org>
- scikit-learn documentation: <https://scikit-learn.org/stable/>
- TensorFlow / Keras documentation: <https://www.tensorflow.org/api_docs>
- PyTorch documentation: <https://pytorch.org/docs/stable/>
- Kaggle — Datasets and ML competitions: <https://www.kaggle.com>
- Papers With Code — State of the art in ML: <https://paperswithcode.com>
Supports
Slides, guided exercises sheets, and practical work sheets.