Mathieu Godbout

Mathieu Godbout, Audrey Durand

Recent work has shown that dynamic programming (DP) methods for finding static CVaR-optimal policies in Markov Decision Processes (MDPs) can fail when based on the dual formulation, yet the root cause for the failure has remained unclear. We expand on these findings by shifting focus from policy optimization to the seemingly simpler task of policy evaluation. We show that evaluating the static CVaR of a given policy can be framed as two distinct minimization problems. For their solutions to match, a set of ``risk-assignment consistency constraints'' must be satisfied, and we demonstrate that the intersection of the constraints being empty is the source of previously observed evaluation errors. Quantifying the evaluation error as the CVaR evaluation gap, we then demonstrate that the issues observed when optimizing over the dual-based CVaR DP are explained by the returned policy having a non-zero CVaR evaluation gap. We then leverage our proposed risk-assignment perspective to prove that the search for a single, uniformly optimal policy via on the dual CVaR decomposition is fundamentally limited, identifying an MDP where no single policy can be optimal across all initial risk levels.

M. Godbout, A. Lachance, F. Antaki et al.

We investigate the potential of machine learning models for the prediction of visual improvement after macular hole surgery from preoperative data (retinal images and clinical features). Collecting our own data for the task, we end up with only 121 total samples, putting our work in the very limited data regime. We explore a variety of deep learning methods for limited data to train deep computer vision models, finding that all tested deep vision models are outperformed by a simple regression model on the clinical features. We believe this is compelling evidence of the extreme difficulty of using deep learning on very limited data.

Nicolas Garneau, Mathieu Godbout, David Beauchemin et al.

In this paper, we reproduce the experiments of Artetxe et al. (2018b) regarding the robust self-learning method for fully unsupervised cross-lingual mappings of word embeddings. We show that the reproduction of their method is indeed feasible with some minor assumptions. We further investigate the robustness of their model by introducing four new languages that are less similar to English than the ones proposed by the original paper. In order to assess the stability of their model, we also conduct a grid search over sensible hyperparameters. We then propose key recommendations applicable to any research project in order to deliver fully reproducible research.