IVADO Workshop

Partagez l'événement


The Fin-ML/IVADO Online Workshop is a practical training in machine learning, applied to concrete problems in investments, banking and insurance. Over each week, the workshop will consist of theory in one session, followed by a problem-solving session. There may be work to complete between the sessions. This online workshop will take place over a 10 week period, 1.5 hour per session with a 15 minute break.

Target Audience

Master’s and PhD students and professionals with quantitative training who want to learn about new technologies in data science, machine learning and operations research, with examples applied to problems in investments, banking and insurance.

*The concepts may be interesting to technical professionals that are not from the finance world, although examples and datasets will come from this environment.


Knowledge of Mathematics & Statistics (Linear Algebra, Probability and Information Theory, Numerical Computation), and Programming (ideally Python) is strongly recommended. The participants can familiarize themselves with Chapters 2 to 4 of the Deep Learning book by Ian Goodfellow and Yoshua Bengio and Aaron Courville (available online deeplearningbook.org)


  • Introduce students and professionals to new technologies in machine learning, that are most relevant in the financial sector;
  • Develop an understanding of the challenges and issues of data science applied to investments, banking and insurance;
  • Learn to use computer tools to solve concrete problems;
  • Encourage interdisciplinary knowledge sharing.


  • Manuel Morales, Associate Professor, Université de Montréal, Director of Fin-ML
  • Rheia Khalaf, Director of Collaborative Research & Partnerships, Fin-ML
  • Nathalie Sanon, Head of IVADO training program


Theme 1: Introduction to machine learning

Week 1 

Theory: Cédric Poutré

  • Types of learning: supervised, unsupervised and reinforced
  • Introduction
    • Regression: Machine learning models vs. statistical approaches
  • Good practices: Overfitting and regularization
  • Good practices: Experimental design

Tutorial: Didier Chételat

  • Exploratory data analysis of an insurance dataset
  • Regression models
  • Generalisation and regularisation

Week 2 

Theory: Cédric Poutré

  • Introduction: Classification
    • Traditional approaches: SVM, Random Forests, etc
    • Modern approaches: neural networks
  • Introduction: Clustering
  • Traditional approaches: K-means

Tutorial: Didier Chételat

  • Classification models
  • Unsupervised analysis

Theme 2 - Introduction to deep learning in investments, banking and insurance

Week 3

Theory: Alexandre Nguyen

  • Introduction: Forecasting problem
  • Multilayered Perceptron
  • Introduction to Deep Neural Network Architectures (CNN, AE)

Tutorial: Aden Houssein Aboubaker

  • Comparison between ML and DL applied in Banking (Fraud Detection):
    • Fraud detection dataset: we will do following steps: Exploratory Data Analysis, Data Preprocessing, Model Evaluation and Selection based on good metric adapted to the situation,
    • We will focus on methodological approaches to handle efficiently the problem of unbalanced datasets in a binary classification context for Fraud Detection,
    • Indeed, the problem of unbalanced data is very frequent in the industry and this is why it is now necessary to have in its Data Scientist toolbox the techniques that allow to remedy this,
    • We will see that there are several solutions to remedy this imbalance problem such as: (1.a) choosing the right metric to monitor, (1.b) solutions at the algorithmic level (e.g. introducing a penalty in the cost function, cost-sensitive learning…), (2) selecting a good threshold (e.g. by optimizing threshold using precision/recall graphs), (3) solutions at the data level (under/oversampling, synthetic data…),, and (4) anomaly detection (e.g. Auto-encoders, Isolation Forest…).

Week 4

Tutorial: Aden Houssein Aboubaker

  • Deep / Representation Learning applied in Insurance (Claim forecasting):
    • How representation learning could allow to boost models
    • How to correctly apply neural network in context of mixed tabular data

Theory: Marie-Ève Malette

  • Generative models
    • Variational Auto-encoders
    • Adversarial networks (GAN)

Week 5 (1 session)

Tutorial: Marie-Ève Malette

  • Conditional variational autoencoders for event occurrence probability estimation
  • Variational Autoencoders in Insurance

Theme 3 - Machine learning in time series forecasting

Week 5 (1 session)

Theory: Alex Nguyen

  • Introduction to recurrent neural networks (continued) (RNN, LSTM, GRU)
  • Forecasting problem

Week 6

Tutorial: Jonathan Guymont

  • LSTM, GRU for times series prediction
  • RNN for times series prediction
  • Use case 1

Tutorial: Jonathan Guymont

  • LSTM, GRU for times series prediction
  • RNN for times series prediction
  • Use case 2

Theme 4 - Introduction to Natural Language Processing in investments, banking and insurance

Week 7

Theory: Alexandre Nguyen

  • Introduction to classical and modern NLP
  • Text processing
  • Language representations

Tutorial: Jonathan Guymont

  • Sentiment analysis
    • Application of convolutional neural networks (CNN)

Week 8 

Theory: Alexandre Nguyen

  • Advanced language models
  • Recent development
  • Applications

Tutorial: Jonathan Guymont

  • Anomaly detection
    • Generative models

Theme 5- Reinforcement Learning in investments, banking and insurance

Week 9

Theory: Alexandre Carbonneau

  • Introduction: learning by reinforcement
    • What is the ‘reinforcement learning problem’?
    • One popular reinforcement learning approach: Q-learning

Tutorial: Didier Chételat

  • Presentation of a financial trading environment
  • Tabular Q-learning

Week 10

Theory: Alexandre Carbonneau

  • Reinforcement learning for large-scale problems
    • Pitfalls of tabular methods of reinforcement learning
    • Function approximators for value and action-value functions
    • Introduction to deep Q-networks
    • Real-life applications of deep reinforcement learning

Tutorial: Didier Chételat

  • Q-learning with linear function approximation
  • Experience replay