David Alvarez-Melis

关于

I am a researcher at Microsoft Research New England in Cambridge, MA. My research revolves around the interplay of statistical, optimization, and geometric aspects of machine learning. The central goal of my work is to make machine learning:

1. more widely applicable, in particular to domains with scarce data, ambiguous objectives, or otherwise constrained; and
2. more trustworthy, by making it more robust, and interpretable

My approach to (1) is ‘data-centric’ (as opposed to model-centric), casting these problems as alignment, comparison, and/or optimization of datasets. For this, my work heavily utilizes optimal transport (opens in new tab), a mathematical gem at the intersection of statistics and optimization.

As for (2), my work seeks to build explanation into machine learning models, and to bring these explanations closer to the way humans explain. For this, I seek inspiration from the study of explanation in the social sciences, and borrow various formal notions of sufficiency, stability, and robustness to formalize it.

In terms of applications, I am particularly interested in problem settings that involve learning with highly-structured data, such as those arising in natural language processing and the natural sciences.

Before starting at MSRNE, he completed a PhD in Computer Science at MIT, and MS and BSc degrees in Mathematics at Courant Institute NYU and ITAM.