In this work, we are exploring the strength of deep learning approaches to learn mid and high level image representations for diabetic retinopathy detection. This project is under development.
The project involves the implementation of a bag-of-words model to classify scences. In the feature extraction was used SIFT and Tiny-Images. The database consisted of 15 different scene categories. The visual vocabulary was built of the training data and gruped with “K-means” and the classification used a linear SVM. The result of the algortihm was summarised in a confussion matrix.
This project seeks to reveal the hidden connections among the most demanded careers in the Peruvian market in order to propose the integration of specialization programs in the Peruvian universities. This way, future professionals will be more competitive. It has been used probabilistic models of topic estimation (Latent Dirichlet Allocation) in text of job ads that was obtained from well known peruvian job websites.
A visualizer of the Peruvian labor demand was implemented between february and july of 2014. This system allows to visualize interactively the individual demand by career, as well as the interrelation between these, that is, when they ask for more than one career in a job ad. For this, each ad was multi-classified in the considered careers using Regular Expressions. You can click here to see the prototype
The approach proposed was a time series regression model build with SVM in order to recognize patterns to make a electric load forecasting model using a public database through Kaggle. This work presented a paper in the Student Project Contest of IEEE Intercon conference in 2013.
The project involves the development of a system capable of modeling companies based on theirs financial statements and estimate the performance that will have in the following quarters, besides determine the variables that affect positively or negatively on that performance. The project is under development, It is been used unsupervised learning algorithms due to the nature of the data.
The project was presented as part of the “XII LARC - Latin American Robotics Competition” in 2013. For this competition, an autonomous robot capable of collect solid trash(represented by cans) on a circular sand surface was developed. The navegation and vision algorithms was operated in real time from a laptop