Ayoub Ghriss

Ayoub Ghriss

PhD Candidate in Machine Learning

PhD in Machine Learning at the University of Colorado Boulder, supervised by Claire Monteleoni. Graduated from Ecole Polytechnique (Computational Mathematics) and ENS Paris-Saclay (MVA: Mathematics, Computer Vision, and Learning).


My research focuses on differentiable graph partitioning and unsupervised learning. I develop theory and GPU-accelerated algorithms for probabilistic graph cuts, connecting spectral methods to modern representation learning. Previously worked on hierarchical reinforcement learning at RIKEN AIP, Tokyo.

CV Open PDF • TeX Source

Research

S$^3$: Structured Sparsity Specification
Ayoub Ghriss — Preprint, Arxiv:2604.11315
Paper • Code • Docs
Beyond Spectral Clustering: Probabilistic Cuts for Differentiable Graph Partitioning
Ayoub Ghriss — AISTATS 2026
Paper • Code • Docs
Crowdsourced Information Authentication: A Graph-based Model from the Science of Hadith
Ayoub Ghriss — Muslims in ML Workshop, ICML 2025
Deep Clustering via Probabilistic Ratio-Cut Optimization
Ayoub Ghriss, Claire Monteleoni — AISTATS 2025
Paper • Code
Predicting Serious Injury and Fatality Exposure using Machine Learning in Construction Projects
E.D. Oguz Erkal, M.R. Hallowell, A. Ghriss, S. Bhandari — J. Construction Engineering and Management, 2024
Paper
Sentiment-aware Automatic Speech Recognition Pre-training for Enhanced Speech Emotion Recognition
A. Ghriss, B. Yang, V. Rozgic, E. Shriberg, C. Wang — ICASSP 2022
Paper
Insights from an Autism Imaging Biomarker Challenge: Promises and Threats to Biomarker Discovery
N. Traut et al. (incl. A. Ghriss) — NeuroImage, 2022
Paper
Deep Learning on Satellite Imagery For Air Quality Modeling
K. Klein, A. Ghriss — Google SWE PhD Internship, 2020
Code
Reinforcement Learning with Options and State Representation
A. Ghriss, M. Sugiyama — RIKEN AIP, 2018
Paper

Software

PGCuts

Differentiable graph cuts with GPU-accelerated hypergeometric kernels. sklearn-compatible API.

Code • Docs

SparseKit

Structured sparsity specification (S3) kit: carve LLM weights the way you wish.

Code • Docs

Sphedron

Refinable polyhedral meshes on the sphere for GNN-based geospatial ML.

Code • Docs

Qwantize

Optimal quantization for block-scaled low-precision formats (NVFP4, MXFP4).

Code • Docs

Tritonix

Triton GPU kernels with an optimized autotuning framework.

Code

Embedata

Dataset and embedding loader for reproducible ML experiments.

Code