Statistics & Data Science

GabrielMoryoussef

PhD in Statistics & Data Science, Carnegie Mellon University
Incoming Member of Research Staff, the Voleon Group

I develop statistical methods for high-dimensional data in which the signal is rare and weak, or hidden in latent structure.

01

About

I recently finished my PhD in Statistics at Carnegie Mellon University, where I worked with Jiashun Jin and Tracy Ke. I spent most of my PhD looking for things that don’t particularly want to be found: signals that are rare, weak, or hiding behind some complicated structure. Lately, I’ve also been getting into large language models, which have plenty of mysteries of their own.

My journey into statistics took a few detours. I started out in pure maths at Imperial College London and Oxford, where I quickly developed a passion for Algebraic Topology and Topological Data Analysis. This led me to write a master’s thesis on the persistent homology of knotted proteins, which is a fancy way of saying I spent a lot of time thinking about knots in things too small to see. Eventually, I decided to transition to the slightly more practical world of machine learning and moved to Paris for the MVA at ENS Paris-Saclay.

After a few years in Pittsburgh, I’m swapping academia for industry (and hopefully better weather): in January 2027, I’ll be joining The Voleon Group as a Member of Research Staff.

ThesisStatistical Methods for High-Dimensional Data with Rare Signals and Latent Structure

Gabriel smiling in a cap and sunglasses at a wooden railing, above a granite creek with mountains behind.
02

Research

  1. i.

    Rare & weak signals

    Higher Criticism and its extensions: detecting and exploiting a vanishing fraction of useful features, with sharp phase transitions for what is and is not possible.

    1 paper
  2. ii.

    Latent structure

    Vertex hunting on the simplex and topic models: recovering the extreme points of a noisy simplex, with improved algorithms and error bounds for successive projection.

    1 paper
  3. iii.

    Ranking & pairwise comparisons

    Bradley–Terry–Luce models that are aware of topics, for ranking journals and large language models from pairwise comparisons.

    1 paper
  4. iv.

    Statistics of LLMs

    Statistical approaches to detecting AI-generated and AI-edited text, and principled comparisons between large language models.

    3 papers
03

Selected publications

NeurIPS 2026Accepted

Multiclass Classification with Rare Useful Features: Fundamental Limits and the Optimality of Diversity Pursuit Higher Criticism Thresholding

G. Moryoussef, A. Kipnis

Advances in Neural Information Processing Systems (NeurIPS), 2026

PDF arXiv

We consider classification in high-dimensional feature spaces where only a vanishing fraction of features are useful and the number of classes may be large. While this rare-and-weak useful-feature regime is well understood in the binary setting, its multiclass version raises new questions: a feature can be useful for class discrimination without separating any specific pair of classes. We focus on a rare feature diversity model, in which most features are useless, but a small fraction promote discrimination among classes in a way that can be detected by a per-feature diversity test, such as one-way ANOVA or the Kruskal–Wallis test. For this setting, we propose Diversity Pursuit Higher Criticism Thresholding (DP-HCT), a feature-screening procedure that applies a higher-criticism threshold to the collection of p-values obtained from per-feature diversity tests. We justify DP-HCT under a Gaussian multiclass Rare-and-Weak Useful Feature model with class-mean contrasts of fixed magnitude in random directions, extending the binary model of Donoho and Jin (2008). When the number of classes is fixed, our model recovers the binary phase transition. When the number of classes grows logarithmically with the number of features, however, we derive a new and strictly harder phase transition, driven by the growing degrees of freedom and their effect on the useful-feature tail rate. We show that whenever classification beyond chance is possible, DP-HCT with ANOVA, followed by a simple classifier based on similarity to class means, achieves asymptotically perfect accuracy. Experiments on real multiclass classification benchmarks demonstrate substantial improvements, and numerical simulations confirm that the derived phase-transition curve accurately predicts the finite-sample behavior of DP-HCT in the growing-class regime.

JASAUnder review

A Topic-Aware Bradley–Terry–Luce Model for Ranking Journals and Large Language Models

Z. T. Ke, G. Moryoussef, Y. Tao, J. Jin

Under review at the Journal of the American Statistical Association (JASA), 2026

Code
Statistical ScienceUnder review

When Machines Write: A Method for Detecting AI-Edited Text

J. Jin, Z. T. Ke, G. Moryoussef

Under review at Statistical Science, 2025

Code
ICLR 2024

Improved Algorithm and Bounds for Successive Projection

J. Jin, Z. T. Ke, G. Moryoussef, J. Tang, J. Wang

International Conference on Learning Representations (ICLR), 2024

Given a K-vertex simplex in a d-dimensional space, suppose we measure n points on the simplex with noise (hence, some of the observed points fall outside the simplex). Vertex hunting is the problem of estimating the K vertices of the simplex. A popular vertex hunting algorithm is successive projection algorithm (SPA). However, SPA is observed to perform unsatisfactorily under strong noise or outliers. We propose pseudo-point SPA (pp-SPA). It uses a projection step and a denoise step to generate pseudo-points and feed them into SPA for vertex hunting. We derive error bounds for pp-SPA, leveraging on extreme value theory of (possibly) high-dimensional random vectors. The results suggest that pp-SPA has faster rates and better numerical performances than SPA. Our analysis includes an improved non-asymptotic bound for the original SPA, which is of independent interest.

All publications (6)
04

News

  1. Upcoming

    Joining the Voleon Group as a Member of Research Staff.

  2. Upcoming

    Presenting our multiclass Higher Criticism paper at NeurIPS 2026 in Sydney.

  3. Defended my PhD thesis, “Statistical Methods for High-Dimensional Data with Rare Signals and Latent Structure.”

  4. “Multiclass Classification with Rare Useful Features” accepted at NeurIPS 2026.

  5. Quantitative research internship at the Voleon Group, Berkeley.

05

Education

  1. 2022 – 2026

    PhD, Statistics & Data Science

    Carnegie Mellon University

    Advised by Jiashun Jin & Tracy Ke. Graduating December 2026.

  2. 2021 – 2022

    MSc, Mathematics, Vision, Learning (MVA)

    ENS Paris-Saclay

  3. 2020 – 2021

    MSc, Mathematical Sciences

    University of Oxford

    Thesis: Persistent Homology of Knotted Proteins.

  4. 2017 – 2020

    BSc, Mathematics

    Imperial College London

    First Class Honours.

06

Contact

Curious about research at Voleon? I’m always happy to chat.

Write to megmoryous [at] andrew [dot] cmu [dot] edu
Based in
Berkeley, CA