Software

Missing values

Many works of this research group have led to actionable implementations for imputation, estimation, and prediction with incomplete data that are mostly available in R.

For a comprehensive overview of missing values problems and methods in R and in python we refer to our platform R-miss-tastic (established with Nathalie Vialaneix and Nicholas Tierney).
Related to this platform is the CRAN Task View on Missing Data maintained by members of this group and Nathalie Vialaneix.

R-packages developed by former and current group members:

  • missMDA: Handling Missing Values with Multivariate Data Analysis (authors: François Husson, Julie Josse)
  • mimi: Main Effects and Interactions in Mixed and Incomplete Data (author: Geneviève Robin)
  • lori: Imputation of High-Dimensional Count Data using Side Information (author: Geneviève Robin)
  • misaem: Linear Regression and Logistic Regression with Missing Covariates (author: Wei Jiang)

Other implementations developed inside our group:

  • PPCA_MNAR (R code): Estimation and imputation in Probabilistic Principal Component Analysis with Missing Not At Random data (author: Aude Sportisse)
  • SGD-NA (Python code): Debiasing Stochastic Gradient Descent with Missing Completely At Random data (author: Aude Sportisse)

Causal inference

Several implementations have been proposed by members of this group to tackle treatment effect estimation with incomplete attributes and on combined experimental and observational data:

  • causal-inference-missing (R code): Doubly robust treatment effect estimation with missing attributes (author: Imke Mayer)
  • combine-rct-rwd (R code): Causal inference methods for combining randomized controlled trials and observational studies (authors: Bénédicte Colnet, Imke Mayer)

Additionally, we have created the CRAN Task View on Causal Inference (maintainers: Pan Zhao, Imke Mayer, et al.). It provides an overview of implementations of causal inference and causal discovery methods currently available on CRAN (The Comprehensive R Archive Network). If you are intersted in contributing or have feedback on this task view, please reach out to the task view maintainers.

Uncertainty quantification

Numerous endeavors from this research group have resulted in practical applications for uncertainty quantification techniques, including the development of methodologies in conformal prediction.

Health applications

Application for bed allocation monitoring: ICUBAM

ICUBAM provides real-time monitoring of intensive care unit (ICU) bed availability in French hospitals. Data is directly obtained from doctors working inside ICU by sending them SMS with a HTTP link to a form that they can fill in 15 seconds.
The project was co-built by ICU Doctors from CHRU Nancy/Université de Loraine and engineers from INRIA & Polytechnique. It was fleshed out live during the Covid crisis in Eastern France to answer an urgent need for finding available ICU beds in a saturated and deteriorating situation. At the time of writing, 5 engineers are working full-time, 7 days a week, on the project, in direct contact with the team of ICU doctors on the ground.