Do Ngoc Tuan (Mason Do)

Research

The main questions I work on, and the papers behind them.

Efficient learning systems

Models and training methods that reach the same accuracy with less compute, memory and data, with guarantees on what is lost.

Related papers (2) →

Efficient Transformers with $\mathcal{O}(n)$ Attention · ICML 2026
Sparse Mixture Models for Low-Resource Learning · AAAI 2025

Optimization theory for machine learning

Convergence guarantees for stochastic and non-convex optimization, and how they explain what works in practice.

Related papers (1) →

On the Convergence of Stochastic Gradient Methods in $\mathbb{R}^n$ under Polyak–Łojasiewicz Conditions · Journal of Machine Learning Research 2025