I study when relational information between individuals can improve machine learning on small, heterogeneous healthcare datasets — and how to evaluate those models reliably.
Machine learning across biomedical engineering, neuroscience, and real-world systems.
I am a PhD candidate at the Institute of Mathematics and Computer Sciences, University of São Paulo (ICMC-USP). My current research focuses on relational machine learning and graph neural networks for heterogeneous Alzheimer-related data, with particular interest in population similarity graphs and inductive inference for unseen individuals.
My background combines Biomedical Engineering, computational neuroscience, and industry machine learning. I conducted fMRI research at the University of Bristol and later worked as a Junior Data Scientist at unico IDTech, where I developed and evaluated machine-learning systems under class imbalance, data scarcity, and production constraints.
Population similarity graphs for reliable learning from heterogeneous clinical data.
My doctoral work investigates whether relational inductive biases can improve cognitive-status modeling when datasets are small, heterogeneous, and moderately imbalanced. The central idea is to represent participants not only by their individual features, but also by their similarity relationships to other participants.
Clinical and sociodemographic participant variables.
Participant-level latent representations from heterogeneous inputs.
Participants connected according to learned or defined similarity.
Relational information propagated with graph neural networks.
Cognitive-status prediction under leakage-controlled validation.
I compare graph-based learning with models that treat participants independently to determine when population structure provides a measurable advantage.
I investigate inductive inference protocols that connect unseen individuals to the learned population structure without rebuilding the full training system.
I use leakage-controlled nested validation, robustness analysis, class-specific evaluation, and ablation studies to separate promising patterns from unstable results.
Open and reproducible work spanning graph learning, neuroimaging, and interpretable machine learning.
A from-scratch implementation of graph neural network foundations, including automatic differentiation, graph operations, GCN and GAT models, and reproducible experiments on Cora.
View repository
Reconstruction and modernization of earlier neuroimaging research on correlation methods and functional connectivity, with an emphasis on transparent, reproducible analysis.
View repository
Reusable Python implementations of model-agnostic interpretability methods, including Ceteris Paribus and Individual Conditional Expectation analyses across multiple datasets.
View repositoryA path from biomedical engineering to relational machine learning.
Advisor: Prof. André Carlos Ponce de Leon Ferreira de Carvalho
Research on relational integration of heterogeneous Alzheimer-related data using population similarity graphs, graph neural networks, rigorous evaluation, and inductive inference.
Developed and evaluated machine-learning systems for large-scale identity verification, working with class imbalance, data scarcity, latency constraints, and GAN-based synthetic-data generation.
Supervisor: Prof. Naoki Masuda
Investigated correlation methods in fMRI data from individuals with autism spectrum disorder, focusing on functional relationships and brain-connectivity patterns.
International academic mobility during the Biomedical Engineering degree, supported by CAPES/BRAFITEC.
Undergraduate training in biomedical engineering, signal processing, and neuroimaging, including research on correlation methods applied to functional MRI.
Early work across biomedical engineering, signal processing, and mathematical modeling.
XII Simpósio de Engenharia Biomédica / IX Simpósio de Instrumentação e Imagens Médicas.
DOIXI Simpósio de Engenharia Biomédica.
DOIXI Simpósio de Engenharia Biomédica.
DOIV Congresso Brasileiro de Eletromiografia e Cinesiologia / X Simpósio de Engenharia Biomédica.
DOIXVII Semana da Matemática / VII Semana da Estatística, Federal University of Uberlândia.
Additional conference work and abstracts are available through my research profiles.
PyTorch · PyTorch Geometric · scikit-learn · XGBoost
Python · MATLAB · C++ · NumPy · Pandas · NetworkX
Nested cross-validation · Robustness analysis · Ablations · Model explainability
Community involvement, recognition, and languages that complement my academic profile.
Volunteer Project Assistant at Data Girls (2023–2025), supporting outreach and mentorship initiatives for women in technology.
Winner of the 2017 Innovations in Electromyography & Kinesiology competition; OBMEP bronze medalist (2010, 2015) with three honorable mentions.
Portuguese (Native) · English (B2) · French (B1)
Research collaborations and academic inquiries.