Research Engineer, Télécom Paris · Institut Polytechnique de Paris
Incoming PhD student, CentraleSupélec
I work on machine learning for graphs and higher-order networks. I am currently a Research Engineer at Télécom Paris, Institut Polytechnique de Paris, working with Jhony H. Giraldo (Télécom Paris) and Fragkiskos Malliaros (CentraleSupélec). I will soon start my PhD at CentraleSupélec under their co-supervision, funded by the DIM AI4IDF, on generative modeling over product graphs for multi-domain (tensor) data.
My current work focuses on inferring latent hypergraphs of group interactions from data without relation labels and, in parallel, on faster Graph Fourier Transforms using low-rank representations and one- and two-step rank updates.
Before that, I was an AI Engineer at Hermes Trade, working on embedding models for retrieval. I received my MSc in Artificial Intelligence and Robotics from Sapienza University of Rome, where I worked on sheaf neural networks for directed hypergraphs with the RSTLess group (published at ICLR 2026) and on ground-to-aerial image matching for visual misinformation detection with the ALCOR Lab.
Started as a Research Engineer at Télécom Paris, working with Jhony H. Giraldo and Fragkiskos Malliaros.
My PhD project with Fragkiskos Malliaros and Jhony H. Giraldo, on generative modeling over product graphs, was selected for funding by the DIM AI4IDF, one of seven projects in the 2026 call. Announcement
Directional Sheaf Hypergraph Networks appeared at ICLR 2026 in Rio de Janeiro.
Joined Hermes Trade in Florence as an AI Engineer.
Graduated from Sapienza with an MSc in Artificial Intelligence and Robotics, 110/110 cum laude.
Our paper on ground-to-aerial image matching was accepted at the AI4MFDD workshop at WACV 2025.
International Conference on Learning Representations (ICLR), 2026
A sheaf-based framework for directed hypergraphs built on the Directed Sheaf Hypergraph Laplacian, a complex-valued operator that generalizes many existing graph and hypergraph Laplacians. Relative accuracy gains of 2–20% over 13 baselines on 7 real-world datasets.
@inproceedings{mule2026directional,
title = {Directional Sheaf Hypergraph Networks: Unifying Learning on Directed and Undirected Hypergraphs},
author = {Emanuele Mule and Stefano Fiorini and Antonio Purificato and Federico Siciliano and Stefano Coniglio and Fabrizio Silvestri},
booktitle = {The Fourteenth International Conference on Learning Representations},
year = {2026},
url = {https://openreview.net/forum?id=d1hQkcDy7c}
}
WACV 2025 Workshop on AI for Multimedia Forensics & Disinformation Detection (AI4MFDD)
SAN-QUAD, a four-stream Siamese-like architecture that matches ground-view images to satellite imagery using semantic segmentation of both views. Improvements of up to 9.8% over prior methods across fields of view on CVUSA.
@InProceedings{Mule_2025_WACV,
author = {Mule, Emanuele and Pannacci, Matteo and Goudarzi, Ali Ghasemi and Pro, Francesco and Papa, Lorenzo and Maiano, Luca and Amerini, Irene},
title = {Enhancing Ground-to-Aerial Image Matching for Visual Misinformation Detection Using Semantic Segmentation},
booktitle = {Proceedings of the Winter Conference on Applications of Computer Vision (WACV) Workshops},
month = {February},
year = {2025},
pages = {795-803}
}
* Equal contribution. See also Google Scholar.
Télécom Paris, Institut Polytechnique de Paris · Paris, France
Hermes Trade · Florence, Italy
RSTLess, Sapienza University of Rome · Rome, Italy
ALCOR Lab, Sapienza University of Rome · Rome, Italy
Sapienza University of Rome · Rome, Italy
CentraleSupélec · Gif-sur-Yvette, France
Sapienza University of Rome · Rome, Italy
Sapienza University of Rome · Rome, Italy