Teaching

”اجعل تعليمك دراسة لعلمك، واجعل مناظرة المتعلم تنبيها على ما ليس عندك.“
— الخليل بن أحمد الفراهيدي
“Homines dum docent discunt.”
— Seneca

Machine Learning

Master 2 SMSDS · Paris Cité University & Sorbonne University

Syllabus

Objectives

Understand the statistical foundations of supervised and unsupervised learning, and implement the main classification and clustering methods in Python or R, with rigorous model evaluation and selection.

Prerequisites

Probability and statistics, linear algebra, and basic programming in Python or R.

Practice

Hands-on exercises in Python or R on real datasets, applying both supervised and unsupervised learning methods.

Outline

  1. Introduction: artificial intelligence, symbolic artificial intelligence and machine learning
  2. Foundations of learning: empirical risk minimization, overfitting, the No-Free-Lunch theorem, the bias–complexity trade-off, structural risk minimization, regularized empirical risk minimization, validation and cross-validation
  3. Supervised classification: k-nearest neighbours as a first algorithm, and evaluation metrics such as accuracy, the confusion matrix, precision, recall, the F1 score, the receiver operating characteristic curve and the area under the curve, the Brier score and Cohen’s kappa
  4. Generative methods: the Bayes classifier, naive Bayes and Gaussian naive Bayes, linear and quadratic discriminant analysis, Gaussian mixture models
  5. Discriminative methods: logistic regression, decision trees and support vector machines
  6. Unsupervised learning: k-means, hierarchical clustering, Gaussian mixture models, DBSCAN and spectral clustering

Slides

Machine Learning, slide 1
Machine Learning, slide 2
Machine Learning, slide 3

Deep Learning

Master 2 SMSDS · Paris Cité University & Sorbonne University

Syllabus

Objectives

Understand how deep neural networks learn representations from data, and design, train and evaluate them for prediction, reconstruction and image analysis.

Prerequisites

The Machine Learning course, with basic linear algebra and calculus.

Practice

Exercises worked by hand and a practical case study on an imaging project.

Outline

  1. Learned representations: artificial neurons, activation functions and the need for non-linearity
  2. Multilayer perceptrons: forward propagation, loss functions, gradient descent and backpropagation, training and overfitting
  3. Autoencoders: encoder and decoder, reconstruction error, denoising and anomaly detection
  4. Convolutional neural networks: convolution, pooling, receptive field, transfer learning, classification and segmentation
  5. Choosing the right model family and building a credible evaluation protocol

Slides

Deep Learning, slide 1
Deep Learning, slide 2
Deep Learning, slide 3

Probability

Master 1 Public Health · Paris Cité University

Syllabus

Objectives

Acquire the foundations of probability theory needed to reason rigorously under uncertainty, with applications in public health.

Prerequisites

Elementary set theory and secondary-school mathematics.

Outline

  1. Introduction: a reflection on chance, and the distinction between epistemic and ontological randomness
  2. Random experiments and Kolmogorov’s axiomatic framework: sample space, sigma-algebra and probability measure
  3. From set-theoretic to probabilistic language
  4. Conditional probability, the law of total probability and Bayes’ theorem
  5. Independence of events and mutual independence

Slides

Probability, slide 1
Probability, slide 2
Probability, slide 3

Statistical Hypothesis Testing

Master 1 Public Health · Paris Cité University

Syllabus

Objectives

Understand the logic of statistical hypothesis testing, from Fisher’s significance tests to the decision framework of Neyman and Pearson, and interpret test results rigorously in public health studies.

Prerequisites

Probability theory, including random variables and their distributions.

Outline

  1. Statistical hypotheses and the purpose of a test: measuring evidence or making a decision
  2. Fisher’s approach: the null hypothesis, the limits of inductive reasoning, test statistics and the p-value
  3. Interpreting the p-value and the significance level, and the limits of Fisher’s approach
  4. The Neyman–Pearson approach: null and alternative hypotheses, simple and composite tests, one-sided and two-sided tests
  5. Type I and type II errors, the power of a test, the rejection region and the critical value

Slides

Statistical Hypothesis Testing, slide 1
Statistical Hypothesis Testing, slide 2
Statistical Hypothesis Testing, slide 3

Artificial Intelligence in Health

Master 1 Public Health · Paris Cité University

Syllabus

Objectives

Understand what artificial intelligence is and how it learns, discover its applications in health, and critically assess its promises and limits.

Prerequisites

None. The course is designed for students without a background in programming.

Outline

  1. What artificial intelligence is: definitions, history and the reasons for its recent acceleration
  2. How machines learn: models, parameters and the learning loop, from machine learning to deep learning
  3. Three ways of learning: supervised, unsupervised and reinforcement learning, illustrated with medical examples
  4. Applications in health: medical imaging, screening, text generation and the prediction of molecular structures
  5. A critical eye: measuring performance, generalizing to other hospitals, and moving from performance to clinical usefulness

Slides

Artificial Intelligence in Health, slide 1
Artificial Intelligence in Health, slide 2
Artificial Intelligence in Health, slide 3