Current Projects

• Google trends for home remedy/medicine relation.
• Introductory workshop for Reinforcement Learning.

Past projects

MSc

• Robust calibration in eye tracking Several models of eye tracking calibration have been proposed, complex models need more information causing delay on in-use calibration. By following the Parsimony Principle -Occam’s Razor- a simple yet accurate and precise calibration is obtained.
  The main idea of this proposal is to compare models using selectors and summary statistics; to accurate determine the best model based on previous techniques. A random sample consensus algorithm is used to ensure robust estimation and evaluation of parameters.
  The best model is selected by person and device; thus, providing personalized selection to a characterized sensor.
  
• Pattern detection in eye tracking Pattern detection is obtain by raw-data eye tracking -differing from common approaches- enhancing accuracy and precision by decomposing eye tracking scanpaths into sub-sequences -q-grams- providing with a least common sub-sequence (LCS) metric.
  In essence, this algorithm uses raw coordinates (x,y) from eye tracking, decomposing onto q-grams or n-sub sequences of q length. Then, compares each query sub-sequence with the equivalence of the pattern creating a robust metric that finds equivalence between a query and a pattern.
  
• Wilderness Search-and-Rescue Paper A multi-agent search-and-rescue algorithm is develop; agents (as a UAV drone simulation) search for a reward (person) in a given environment. This heuristic metric is simulated with given traits of a particular person and UAV; outperforming sweeping algorithms (Spiral, Lawnmower, Uniform Circular Sector).
  This algorithms ensures neither of the agents explore the same territory on the same period of time, most of the area is covered, attraction to high-probability areas; the results show that a reduction in time is achieved with 2-5 agents versus common techniques.
  

BSc

• Smartband A smartband prototype is develop to detect hand gestures using computer vision and sensor information (accelerometer and gyroscope). A basic GUI is design to help the user connect the device with external environments; usage as mouse and hot-keys for Windows is the principal option.
• Self balance 1d system A CAD/CAE fuzzy-controlled 1-degree balance system using ‘low-level’ linear algebra on MATLAB. Connection from MATLAB was made to visualize in real time assembly movements of the CAD real-model on Solidworks; using Labview as intermediary.