Gabriel Melo

Gabriel Melo

PhD Student · Machine Learning · Graphs & Uncertainty

Papers

Flow Figure

MSAlign: Aligning Molecule and Mass Spectra Foundation Models for Metabolite Identification

Paul Krzakala, Gabriel Melo, Camille Lançon, Charlotte Laclau, Rémi Flamary, Etienne Thévenot, Florence d'Alché-Buc
Pre-print, 2026
TL;DR: Introduces MSAlign, a method for aligning molecule and mass spectra foundation models to improve metabolite identification.
ICML
Flow Figure

Learning to Emulate Chaos: Adversarial Optimal Transport Regularization

Gabriel Melo, Leonardo Santiago, Peter Y. Lu
ICML, 2026
TL;DR: Introduces an adversarial optimal transport regularization method to improve the stability and generalization of neural emulators when learning chaotic systems.
ICML

Lightweight Alignment of Unimodal Foundation Models for Metabolite Identification

Paul Krzakala, Gabriel Melo, Camille Lançon, Charlotte Laclau, Rémi Flamary, Etienne Thévenot, Florence d'Alché-Buc
Workshop on Multi-modal Foundation Models and Large Language Models for Life Sciences @ ICML, 2026
TL;DR: Short version of MSAlign.
NeurIPS
Flow Figure

Conformal Graph Prediction with Z-Gromov Wasserstein Distances

Gabriel Melo, Thibaut de Saivre, Anna Calissano, Florence d’Alché-Buc
UAI, 2026
TL;DR: Extends conformal prediction framework for graph-valued outputs using Z-Gromov Wasserstein distances as non-conformity score.
NeurIPS
GRALE Project Figure

The quest for the GRAph Level autoEncoder (GRALE)

Paul Krzakala, Gabriel Melo, Charlotte Laclau, Florence d'Alché-Buc, Rémi Flamary
NeurIPS, 2025
TL;DR: Proposes GRALE, an optimal-transport based autoencoder for graph-level representation learning.
Example Project Figure

Study of Impedance Matching in CPW Cavities for Circuit QED and Quantum Computing Applications (in Portuguese)

G Melo, Francisco Rouxinol
Galoá, 2021
TL;DR: Designs and simulates a Marchand balun to improve signal coupling for superconducting quantum circuits.

Blogposts

Donut Transformer Blogpost

Fine-tuning Donut Transformer for Document Classification

Gabriel Melo
Medium, 2024
TL;DR: Compares encoder-decoder vs encoder-only fine-tuning for Donut; encoder-only is faster and equally accurate for document classification.