I am currently a PhD student at the University of Zaragoza, under the supervision of Elena Garces and Prof. Diego Gutierrez. My main research interests lie in finding the relationship between modern statistical methods and their application to computer graphics problems.
Before starting the PhD I obtained my bachelors and masters degree at the University of Zaragoza, majoring in Computer Science and Applied Maths respectively; at the same time, I was an undergraduate researcher at the Graphics & Imaging Lab, advised by Elena Garces.
PhD in Computer Science, 2017
Universidad de Zaragoza
MSc in Mathematical Modelling, Statistics and Computation, 2016 - 2017
Universidad de Zaragoza
Bachelor in Computer Science, 2012 - 2016
Universidad de Zaragoza
Bachelor in Software Engineering, 2014 - 2015
Aarhus University (exchange programme)
Implementation of a U-net like model used for car segmentation. The code with two purposes in mind: first, to participate in the Carvana Image Segmentation challenge run on Kaggle platform, and second to learn and practice with the Pytorch framework. The model was able to get an accuracy of approximately 0.996. To do so, it combines BCE and Dice loss during training in a standard convolutional U-net. Further improvementes would be to add dense connections to the layers to improve the performance.
Library that aims to be an abstraction of the evolutionary process followed by genetic algorithms. It supports different evolutionary paradigms like Genetic Algorithms (ga), Evolution Strategies (es) and Grid-based Genetic Algorithm for Multimodal Real Function Optimization by Jose Chaquet and Enrique Carmona (gga). It has implemented several mutations, crossovers and selection mutations.
Following project shows an analysis of the data acquired by the marketing time of a bank. The collected data comes from phone calls and the main goal will be to predict if a client will subscribe to some bank deposit and improve efficiency and efficacy by the identification of the core variables that influence the client decision. The data is highly imbalanced, something that is taking into account either by doing weighted, SMOTE, up or down sampling. I tried several methods, including standard decision trees, conditional inference trees, bagging, boosting, and SVMs. (documentation in spanish)
The following project tries to solve a reverse engineering problem which consists on getting the correct answers of an exam given the answers of the students and their grade (none of them guessed all the questions correctly, indeed the best mark was ^{17}⁄_{25}). To solve the problem I use predictive models, in particular I tried a penalized Lasso regression, then by looking at the biggest coeficient for each question I can guess the answer for each question. However, using Lasso regression only 23 questions were guessed correctly. Neither random forest nor bagging or boosting methods improved the Lasso regression. I also tried by changing the missing values for the expected one but it did not help either. Finally, with a genetic algorithm approach I managed to obtain the 25 answers correctly. (documentation in spanish)