My personal web-page on the website of my organization is very outdated and it will only be updated in January due to the holidays. This current web-page serves as my professional page until my organization website is updated.

Bio

Alexia is the Biostatistician of the Dream Big research team at the Child Center for Development and Mental Health. She provides close assistance to the team and act as a co-mentor to the M.Sc. and Ph.D. students at the lab. She is primarily interested in developing new statistical approaches (or improving already existing approaches) to better explain, predict and generate from data.

To better understand how genes interact with the environment to predict how children develop, she devised a novel approach which considers multiple genes and environment into a single model. Traditionally, researchers have mostly focused on looking at one gene at a time. In reality, a single gene is not enough to fully explains how a person will develop or behave. Some researchers have also been using the full genome, but this is problematic for a different reason: it makes it impossible to determine the individual contribution of specific genes when there are so many considered at the same time. Her approach instead focuses on looking at a small selection of genes and environments so that one can better understand how they work together. This approach is called Latent Environmental and Genetic InTeraction (LEGIT) and it is freely available online. This approach was later adapted to be able to not only model the interaction between multiple genes and environments, but also to determine the specific type of gene-by-environment interaction involved: diathesis-stress (i.e., a person is only affected by the negative environments), vantage sensitivity (i.e., a person is only affected by the positive environments), or differential susceptibility (i.e., a person is affected by both the good and bad parts of the environment; for better or worse).

She is also known her research in Artificial Intelligence (AI). Her first experience in AI started with the Meow Generator in which she used variants of an approach called Generative Adversarial Networks (GANs) to generate pictures of cats 🐈. GANs consist in two artificial neural networks (D and G) that fight one another. The discriminator (D) tries to determine which data is real and which is fake, while the generator (G) tries to generate fake data by fooling the discriminator into thinking that fake data is actually real. Although used to generate pictures of cats, GANs can actually be used to replicate any type of data (text, sound, pictures, psychological data, etc.), and thus have many applications. Recently, she improved on GANs by devising a new approach called Relativistic GANs. With this improvement, she was able to train beautiful cats in high resolution (256 x 256), a feat not achievable by the standard approach. This approach has also been widely used by other researchers; Wang et al. (2018) used Relativistic GANs to win a competition in super-resolution (producing high resolution from low resolution images).

Recent Publications

  1. Jolicoeur-Martineau, A. (2018). The relativistic discriminator: a key element missing from standard GAN. arXiv preprint arXiv :1807.00734
  2. Jolicoeur-Martineau, A. (2018). GANs beyond divergence minimization. arXiv preprint arXiv :1809.02145
  3. Graffi, J., Moss, E., Jolicoeur-Martineau, A., Moss, G., Lecompte, V., Pascuzzo, K., Babineau, V., Gordon-Green, C., Mileva-Seitz, V.R., Minde, K., Sassi, R., Steiner, M., Kennedy, J.L., Gaudreau, H., Levitan, R., Meany M.J., & Wazana, A. (2018). The dopamine D4 receptor gene, birth weight, maternal depression, maternal attention, and the prediction of disorganized attachment at 36 months of age: A prospective gene x environment analysis. Infant Behavior and Development, 50, 64-77.
  4. Jolicoeur-Martineau, A., Belsky, J., Székely, E., Widaman, K.F., Pluess, M., Greenwood, C.M., Wazana, A. (In press, 2018). Distinguishing differential susceptibility, diathesis-stress and vantage sensitivity: beyond the single gene and environment model. arXiv preprint arXiv :1712.04058
  5. Jolicoeur-Martineau, A., Wazana, A., Székely, E., Steiner, M., Fleming, A. S., Kennedy, J.L., Meaney, M.J., Greenwood, C.M. (In press, 2018). Alternating optimization for GxE modelling with weighted genetic and environmental scores: examples from the MAVAN study. arXiv preprint arXiv :1703.08111

Google Scholar

GitHub

Advertisement