Mame Diarra Toure

I am a PhD candidate in Applied Mathematics at McGill University, advised by David Stephens. My research focuses on Bayesian Deep Learning and Uncertainty Quantification.

I develop scalable variational methods to make neural networks robust, interpretable, and computationally efficient. My goal is to bridge the gap between theoretical rigor and practical deployment to build safe and trustworthy AI systems.

In 2024, I was awarded the Women in AI Excellence Scholarship from Mila (Quebec AI Institute).

Prior to McGill, I worked as a Quantitative Analyst at Société Générale. My academic foundation was built in France, where I completed the intensive Classes Préparatoires aux Grandes Écoles before earning an Engineering Degree from ENSIIE and an MSc in Quantitative Finance from Paris-Saclay University, where I graduated with First Class Honours and was awarded the Sophie Germain Excellence Scholarship by the Jacques Hadamard Mathematic Foundation(FMJH)

Email  /  CV  /  GitHub

Research & Publications

My primary research focuses on developing scalable low-rank variational inference methods (W ≈ AB^T) for Bayesian neural networks. I also investigate whether these structures allow networks to automatically adapt their complexity to the data (Occam's Razor).

We introduce a low-rank variational inference framework for Bayesian neural networks that parameterizes weights as W = AB^T, reducing parameter complexity from O(mn) to O((m+n)r). This induces a posterior singular with respect to the Lebesgue measure, capturing structured weight correlations through shared latent factors. We derive PAC-Bayes generalization bounds and prove loss bounds using the Eckart-Young-Mirsky theorem. Empirically, our method achieves competitive predictive performance with 5-member Deep Ensembles while using up to 15× fewer parameters and substantially improves out-of-distribution detection across MLPs, LSTMs, and Transformers.

Developing a variational inference framework that factorizes weight matrices to reduce parameter complexity from O(n²) to O(n), enabling uncertainty quantification in large-scale models.

Applied Deep Neural Networks to solve high-dimensional PDEs for Credit Valuation Adjustment (CVA), overcoming the curse of dimensionality in traditional finite difference methods.

Building ANN models to learn option price surfaces from market data.

Implemented statistical techniques to estimate the kernel of Volterra processes, validating the Rough Volatility hypothesis in financial time series.

Experience

TA for MATH 139: Calculus with Precalculus. Leading problem-solving tutorials and strengthening students' conceptual understanding of limits, differentiation, and functions.

TA for MATH 141: Calculus 2. Leading problem-solving tutorials and strengthening students' conceptual understanding of the definite integral, techniques of integration, applications and introduction to sequences and series. .

Contributed to applied R&D projects in NLP and Reinforcement Learning. Designed a collaborative filtering matchmaking algorithm for startups and optimized budget pacing for advertising using RL.

Selected for an exclusive mentorship program providing in-depth exposure to quantitative research methodologies, financial modeling, and career pathways in global markets.

Contributed to the Haussmann Project. Developed and calibrated PD (Probability of Default) and LGD (Loss Given Default) models for low-default portfolios.

Developed mathematical models to quantify the bank's exposure to Central Counterparty Clearing Houses (CCPs). Modeled default fund contributions using Monte Carlo simulations.

Community & Mentorship

I am an active member of Women in AI North America and the Association of Women in Mathematics. I also serve as a mentor at Jiggen In STEM, supporting high school girls in Senegal, and was the former representative of AFSA Québec at McGill.

© 2026 Mame Diarra Toure