4 présentations

• 
  15h30 - 15h55

  Non-parametric choice modeling with product-oriented market segmentation

  • Milad Keshvari Fard, prés., HEC Montréal
  • Sanjay Dominik Jena, Université du Québec à Montréal
  • Laurent Charlin, HEC Montréal

  In this research we propose a non-parametric choice model to improve demand forecasting. Our algorithm first identifies the preference of different market segments for products, followed by computing ranking distributions over preference lists specific to each segment. Our results indicate a significant improvement in demand forecasting as well as computational efficiency over the state of art.
  Keywords: Non-parametric choice modeling; Assortment optimization; Market segmentation; Clustering; Machine Learning.
• 
  15h55 - 16h20

  And-Or decision diagrams for multi-stage decision making under uncertainty

  • Behrouz Babaki, prés., Polytechnique Montreal
  • Golnoosh Farnadi, Polytechnique Montréal
  • Gilles Pesant, Polytechnique Montréal

  Factored stochastic constraint programming (FSCP) is a formalism to
  represent multi-stage decision-making problems under uncertainty. On one
  hand, it relies on methods from constraint programming for making the
  decisions, and on the other hand, it uses principles from uncertainty
  reasoning to deal with a probabilistic environment. However, solving
  these problems is computationally challenging. FSCP problems often
  involve repeated subproblems which ideally should be solved once. In
  this work, we show how identifying and exploiting the identical
  subproblems can simplify solving the FSCP problems and leads to a
  compact representation of the solution.
• 
  16h20 - 16h45

  Verifying individual fairness in machine learning

  • Golnoosh Farnadi, prés., Polytechnique Montréal
  • Behrouz Babaki, Polytechnique Montreal
  • Gilles Pesant, Polytechnique Montréal
  • Michel Gendreau, Polytechnique Montréal

  Nowadays, machine learning tools are used in various decision making services and in domains that influence peoples' lives such as policing, employment, health care, and education. While many may assume that automation removes human bias from decision-making, it has been shown that bias can be part of the design of the algorithms or inherited from data that is used by the algorithm. Due to the large popularity and success of machine learning (ML) in various applications in recent years, in this paper we focus on verifying fairness in popular ML models. We propose an efficient and effective individual fairness verification approach based on MIP/CP formulation.
  Keywords: Fairness, Machine Learning, Verification
• 
  16h45 - 17h10

  Social welfare on kidney transplantation

  • Margarida Carvalho, prés., Université de Montréal
  • Andrea Lodi, Polytechnique Montreal

  Kidney exchange programs when modeled as non-cooperative games between different entities (hospitals, countries, regions) have proven to lead to social optimal outcomes. In this work, we discuss how the game outcome changes according with the data available about patients and donors.