Matt Rodriguez

I am programmer analyst at SDSC in the geospatial group headed by Ilya Zaslavsky. The group provides production web services and GIS applications for CUAHSI . CUAHSI is a consortium of universities centered out of University of Texas trying to organize hydrologic datasets to enable hydrologic science. I've built a GIS web application that queries a data cube and generates maps using the ArcGIS web framework. Map Maker

Currently I work mostly with the Ocean Observatories Initiative project. We are developing an AMQP Messaging system to deliver large amounts of Oceangraphic data. This project I get to work on Amazon's EC2 , use the distributed key value store Cassandra. One implementation of the Messaging system uses Python Twisted. The technology used on this project is current and exciting.

I've worked with Chesapeake Bay Environmental Organization developing web applications that allow them to analyze and visualize their data. The first web site allows the user to search through two datasets and filter by spatial and temporal constraints. The application runs a matching algorithm that matches data points from one data set to the other. It generates a scatterplot of the data and calculates the correlation coefficient. The web application is here.

I made a SQL Server report which calculates the Hypoxic volume and percentage of the volume. Th report executes a stored procedure. The data set is a model data set which contains over 25 billion data points. I aggregrated the data and built indexes to make the procedure execute in a tolerable amount of time. The report can be found here here. The report requires a username and password to run.

I have a Masters in Computer Sciences and Engineering from UCSD. My academic interest focus on machine learning and data mining. My Masters thesis is focused on Recommender Systems. The problem typically reduces to estimating a matrix where rows are represented by users and columns by items. Most of the matrix elements are unknown. I am exploring different techniques for estimating the Recommender System matrix. The latest algorithm is the PDLF algorithm by Deepak Agarwal and Srujana Merugu. Here's a link to a kdd07 video presentation .

My Masters Thesis is found here.


Presentation of the Gravity Recommender System Slides for Charles Elkan's 291 course in Spring 2008.

Presentation of PDLF paper Slides for Lawrence Saul's 291 course in Winter 2009.