Ongoing and completed projects spanning healthcare AI, precision agriculture, complex network theory, and language models.
High-level data classification using network topology, community detection, structural optimization of networks, K-associated optimal graphs, PSO-guided network search, and multi-label classification via complex network measures.
CNN-LSTM hybrids for biomedical signals, Echo State Networks for EEG coma prognosis, transformers for spectroscopy data generation, and convolutional architectures for molecular pattern detection.
Support vector machines, random forests, high-level classification, unsupervised learning, dimensionality reduction with K-associated graphs, semantic role labeling, and ensemble methods.
Synchronous cellular automata for scheduling, bio-inspired network structural optimization, surrogate-assisted genetic algorithms for peptide discovery, and evolutionary computation for hospital management.
Large language and foundation models, transfer learning, and generative approaches for spectral data synthesis and domain-specific fine-tuning.
All ongoing projects funded and running in the AINET laboratory.
Collaboration with TU/e (Eindhoven) and Radboud University (Netherlands). Developing an automated ML platform for detecting Autism Spectrum Disorder from saliva using ATR-FTIR spectroscopy with Fourier-transform infrared analysis. Targets AutoML configurations capable of optimizing pre-processing and classification pipelines for a variety of spectral data modalities.
Development of large language models fine-tuned and evaluated for the finance domain. The project investigates domain-specific pre-training strategies, evaluation benchmarks for financial NLP, and deployment-friendly architectures with reliability and explainability constraints.
Design of hybrid learning architectures for mono and multi-sequence classification using EEG and ATR-FTIR spectroscopy data. Focus on detecting patterns in FTIR and electromyography/electroencephalography data. Collaboration with Prof. Liang Zhao (USP) and Prof. Ran Cheng.
High-level data classification in complex networks for molecular diagnosis of autism spectrum disorder (TEA) from salivary molecular properties analyzed via ATR-FTIR spectroscopy. Investigates structural and dynamic network properties to improve conventional classification methods.
Network-based machine learning models for early prognosis of patients in deep coma states, using EEG data. Investigation of Echo State Networks, graph neural networks, and long short-term memory architectures for early prediction of outcomes, aiding organ donation decisions and clinical management.
Machine learning for agricultural challenges: soybean maturity classification via high-throughput UAV RGB imaging and CNNs; eucalyptus growth and production modelling; coffee harvest time prediction with cellular automata; and general geospatial information processing for Brazilian biomes.
Complex network-based deep learning models for rapid, non-invasive COVID-19 diagnosis from saliva samples processed via near-infrared spectroscopy. Two-time Google Latin America Research Award winner (2020 & 2022). In collaboration with Prof. Robinson Sabino-Silva and Ministry of Health (Brazil).
Development of a scalable technological platform for COVID-19 diagnosis integrating ML-powered spectroscopic analysis, rapid testing workflows, and clinical decision support. Funded by the Brazilian Ministry of Health.
Investigation of methods and heuristics based on complex networks for data classification and clustering tasks. Develops approaches beyond traditional graph metrics — exploring the dynamic and structural properties of data for multi-label and multi-class scenarios.
Nanotechnology-based sustained-release oral liberation system for SARS-CoV-2 prevention. ML-based spectroscopic structure detection for early release triggering.
Identification of the main indicators contributing to student evasion and failure in UFU computing courses using data mining techniques. Resulted in institutional improvement actions and one registered software tool (SAGAPEP).
Intellectual property resulting from AINET research.