My core background is physics but in the last 15 years, my research was been focused on Machine Learning, Artificial Intelligence, Deep Learning Neural Networks and Complex Network analysis.
Since 2000 I have been working and coordinating research projects on machine learning. Most of these projects are related to processing, filtering and classifying high-dimensional data. Some of these techniques and algorithms are
- Artificial Neural Networks;
- Support Vector Machines;
- MOEGA – Multi-objective Evolutionary Genetic Algorithms
- Multidimensional Scaling (MDS);
- Isomap – supervised and unsupervised;
- Non-negative Matrix Factorization (NMF)
These algorithms were applied to Thin Film Analysis with Rutherford Backscattering (RBS), Credit Risk, Bankruptcy Prediction, Data Filtering, Protein Folding, and web recommendations (see publication list and projects).
I developed a new algorithm to train Neural Networks for classification problem. To see its application Financial Distress Prediction of private French companies see project AIRES at Aires-risk.com
With Gaspar Cunha, we developed a method to accelerate the search of the Pareto-front in Multi-Objective Evolutionary Genetic Algorithms with Neural Networks.
One of the largest problems of users and organizations is to retrieve relevant information on large unstructered datasets: Internet, social networks and e-commerce sites. Recommendations systems (RS) are crucial to solve this problem. However, traditional RS rely on Collaborative Filters, which suffer from well known deficiencies, like the “cold start problem”, thus requiring large amounts of user feedback in order to retrieve useful data.
Detection of Opinion Leaders and Spread of Influence on Social Networks
Recently I start working on Complex Network analysis and its applications to recommender systems and information spreading in social networks, namely scale free weighted networks and fuzzy graphs. We are applying these models to document ranking, and social networks analysis – information spreading and network resilience.
MyRec: recommender system based on complex networks
One of the most exciting works I ever did was with Nuno Barradas on optimizing RBS diagnostics for thin films using Neural Networks.Together with Nuno Barradas we have developed a fully automated Rutherford Backscattering data analysis capable to adjust all experimental parameters in order to obtain the best possible spectra for a specific sample to be analyzed.