See my full professional profile at LinkedIn.
academic background |
|||||
Universitat Autonoma de Barcelona (Centro de Visión por Computador) |
|||||
Ph.D. Student in Computer Science | 2015-now | ||||
Computer Vision. Specializing in machine video-based learning. | |||||
Universidade Federal de São Carlos, Brazil |
|||||
Master of Science in Computer Science | 2011-2013 | ||||
Image and Signal Processing. Specialized in vision-based sign language recognition. | |||||
Universidade Federal de São Carlos, Brazil |
|||||
Bachelor in Computer Science | 2007-2011 | ||||
Neural networks for creating computer-aided medical decision support systems. | |||||
professional experience |
|||||
Xerox Research Centre Europe, France |
|||||
Research and Development Engineer in machine learning for Transportation Services. Co-designed a modular pricing optimization framework and created the entire machine learning modelling, implementation and execution necessary for realizing this framework into an actual product. Authored two patents on how to apply machine learning to parking management and simulation models, and co-authored a third patent on methods for transport optimization within large cities. | 2013-2015 | ||||
DaitanGroup, Brazil |
|||||
Systems Development Specialist in high-load, highly-scalable telecommunication and multimedia systems for VoIP-enabled applications. Reworked a start-up project with maintainability in mind; adding flexibility that would cut costs in the short and long run. Later on, used this same architecture to create an application framework with support for multiple network protocols such as SIP and BFCP. | 2013-2013 | ||||
Universidade Federal de São Carlos, Brazil |
|||||
Researcher in Image and Signal Processing for sign language recognition. Investigated how machine learning, pattern recognition, artificial intelligence and statistical analysis could be used to recognize hand movements in depth images. | 2011-2013 | ||||
Daitan Labs, Brazil |
|||||
Software Development Specialist in mission critical telecommunication applications in C/C++ for UNIX environments. Focus on redundancy, fault tolerance and large (countrywide) scalability. SIP stack protocols and telecommunication RFCs. | 2010-2011 | ||||
Universidade Federal de São Carlos, Brazil |
|||||
Undergraduate scholarship researcher in neural networks for the prognostic evaluation of post-surgery complications in patients underwent to myocardial by-pass graft surgery. | 2007-2009 | ||||
skills |
|||||
I have created a machine learning, computer vision and image processing framework called Accord.NET. This has brought me theoretical and practical knowledge on a variety of machine learning and computer vision topics, such as (but not limited to):Mathematics and statistics; linear and non-linear function optimization (gradient-free, constrained and unconstrained), matrix manipulation, decompositions and factorizations (such as non-negative factorization), statistical hypothesis testing (parametric and non-parametric), classical methods and models (linear, generalized linear and non-linear, link functions), data analysis (linear and non-linear component, discriminant, factor and latent analysis), performance assessment (cross-validation, bootstrapping, statistical measures for contingency tables and confusion matrices);
Machine learning and artificial intelligence; sequence classifiers (Markov models, Conditional Random Fields and their latent-variable variants); generative and discriminative models, neural networks (first order and second order optimization methods, first-generation deep learning [based on restricted Boltzmann machines]), support vector machines (multi-class and multi-label strategies, DDAGs, statistical learning theory with main expertise on Kernel Support vector Machines); general scientific computing (research and experimental workflow, parallel processing) and specialized data structures (KD-Trees); Computer Vision and Image processing; feature extraction (HOGs, FREAK, SURF, Haralick, Local Binary Patterns); corners detection (Harris, FAST, Susan); feature matching (kNN, correlation matching), homography estimation; image moments, bag of visual words, RANSAC, computer vision (color and template based object trackers, object detectors, filters, boosted classifiers). |
|||||
selected publications |
|||||
J Myers, CR Souza, A Borghi-Silva et al., A neural network approach to predicting outcomes in heart failure using cardiopulmonary exercise testing, International Journal of Cardiology, 2014 Feb 1;171(2):265-9.
In this paper, published in the International Journal of Cardiology by Elsevier, we described a study on the use of artificial neural networks to predict cardiovascular (CV) death in patients with heart failure (HF). We compared the applicability of relatively modern techniques in the medical field, in comparison with the standard Cox’s proportional hazard models and common regression techniques mostly found in the medical literature. CR Souza, EB Pizzolato, Brazilian Sign Language Recognition with Support Vector Machines and Hidden Conditional Random Fields Machine Learning and Data Mining in Pattern Recognition – MLDM 2013. Lecture Notes in Computer Science Volume 7988, 2013, pp 84-98, Springer Berlin, 2013. In this study, we had shown how to recognize freely articulated sign words using Support Vector Machines and Hidden Conditional Random Fields in the Brazilian Sign Language (Libras). We performed the classification of natural signed words in unconstrained backgrounds without the aid of gloves or wearable tracking devices. We have demonstrated that feature vectors extracted from depth information and based on linguistic investigations could be effective for this task. This paper has been selected as one of the three best papers published in the conference. CR Souza, EB Pizzolato, MS Anjo, Fingerspelling Recognition with Support Vector Machines and Hidden Conditional Random Fields Advances in Artificial Intelligence – IBERAMIA 2012. Lecture Notes in Computer Science Volume 7637, 2012, pp 561-570, Springer Berlin, 2012. In this paper, we investigated how to use Hidden Conditional Random Fields and Support Vector Machines in the problem of fingerspelling recognition in Libras. We achieved much faster evaluation rates using large-margin Directed Acyclic Graphs to perform multi-class classification in SVMs. We provided comparisons against more common approaches based on Artificial Neural Networks and Hidden Markov Models. J Myers, CR Souza, A Borghi-Silva et al., A Neural Network Approach to Predicting Outcomes in Heart Failure using Cardiopulmonary Exercise Testing (abstract) Journal of the American College of Cardiology, v. 61, p. E1415, 2013. March 12, 2013 Short version of the first publication, for the Journal of the American College of Cardiology. S Feuerstack, JH Colnago, CR Souza et al., Designing and Executing Multimodal Interfaces for the Web based on State Chart XML Proceedings of The 3rd W3C Brazil Web Conference, Rio de Janeiro, Brazil In this paper, we had shown how to create and specify multi-modal interfaces using the state-chart XML language specification. In this paper, we created and tested a computer vision based module to control an interface for automatically turning musical sheets using visual gestures. CR Souza, EB Pizzolato, R Mendes et al., Artificial Neural Networks Prognostic Evaluation of Post-Surgery Complications in Patients Underwent To Coronary Artery Bypass Graft Surgery 18th International Conference on Machine Learning and Applications (ICMLA’09), 2009, Miami Beach, FL In this study, we evaluated the use of artificial neural networks in the prediction of post-surgery complications in CABG, such as the need for patient re-intubation, exposure to prolonged mechanical ventilation times and mortality risk increase. |
|||||
honors and awards |
|||||
Scientific computing for Java and Android.
The CodeProject, 2013Prizewinner article in CodeProject’s Android article competition of November 2013. We describe how to use the Accord.NET Framework to build applications for Android, enabling researchers and developers to apply machine learning and image processing techniques in embedded systems. Haar-feature Object Detection in C#. The CodeProject, 2012 Prizewinner article in CodeProject’s C# article competition of August 2012. A reimplementation of the Haar cascade recognition system from Viola & Jones in .NET, compatible with OpenCV’s XML definition files, employing a few techniques for faster recognition and featuring a multi-threaded based implementation. Handwriting Recognition using Kernel Discriminant Analysis. The Code Project, 2010 Double prizewinner article in CodeProject’s C# article competition and best overall article of May 2010. Describes how to use the Kernelized version of Linear Discrimination Analysis to obtain a Non-linear discriminant analysis applicable to handwriting recognition. |
|||||
patents |
|||||
US application 14/632,207. February 26th, 2015Methods and systems for interpretable user behavior profiling in off-street parking.
US application US 14/630,227. February 24th, 2015 Method and system for simulating users in the context of a parking lot based on the automatic learning of a user-choice decision function from historical data considering multiple user behavior profiles. |
|||||
work relevance and impact |
|||||
Over the years, my work has been directly used to empower more than 40 distinct applications, for amateurs, academics and professionals, including paper publications, master theses and PhDs. A non-complete list of those works is available at http://accord-framework.net/publications.html. Some of the most interesting works are showcased here:
|
|||||
languages |
|||||
Portuguese (native), English (bilingual), Spanish (professional), French (professional) |