Oscar Madrid Padilla will become the first person to receive a PhD by the University of Texas at Austin's Department of Statistics and Data Sciences (SDS) this May. The department was formed in August 2014 and replaced the Division of Statistics and Scientific Computation.
Madrid Padilla, a native of Honduras, has done research with broad applications, and often works at the intersection of statistics and other fields, having collaborated with researchers across UT Austin and at other universities. His advisor was James Scott, a professor in SDS and the McCombs School of Business.
Scott and Madrid Padilla collaborated with Alex Athey, of UT Austin's Applied Research Laboratories, on work which could help in the detection of radioactive material in real time, a valuable resource for law enforcement trying to prevent the use of a so-called dirty bomb, where explosives are paired with radioactive material, with potential to unleash a devastating terrorist attack.
There are two key challenges in detecting radioactive anomalies and alerting authorities. First is that the intensity of the anomaly is usually very small compared to the background radiation of the surroundings. And second, the detection of such material must be done in real time.
To overcome those limitations, Madrid Padilla took inspiration from a set of statistical ideas, based on studying so-called stochastic processes. He created computer algorithms which could detect small anomalies in radiation effectively in real time with limited data.
To utilize the new technique, law enforcement would first scan an area to collect baseline radioactivity data. Then, going forward, they could use a simple handheld sensor to monitor the same area. The algorithm would compare the input to the baseline data and respond if there is an anomaly. Existing systems which do this sort of real time detection raise many false alarms, which means wasted resources and less trust in the system. Therefore there is a delicate balance between being conservative enough to not raise false alarms but also being sensitive enough to raise a warning when there really is an anomaly. Madrid Padilla's system is faster and gives fewer false positives than the current alternatives.
Melding mathematics, statistics, and computer science, is a common thread in Madrid Padilla's work.
"My intuition is mathematical," he says, "but just because an idea is beautiful in mathematics doesn't mean that it will work in real life. In computer science there are a lot of limitations in what you can and cannot do. So that's something you have to overcome."
Originally born in Honduras, Madrid Padilla was one of the first two high school students from his country to qualify for the prestigious International Mathematical Olympiad (IMO), where he earned an honorable mention for completing one of the problems perfectly. It was at the IMO that he met a representative from The Center for Mathematical Research (CIMAT) in the Mexican city of Guanajuato. Impressed by his mathematical ability, they offered him a scholarship to study in Mexico.
Even before the IMO competition, Madrid Padilla had deep interests in mathematics thanks to his father, who "wasn't a mathematician, or a statistician, he didn't even go to college, but he always said how important mathematics was, and he would find books on math and read them and then teach the material to me."
He earned his undergraduate degree in pure mathematics at CIMAT. Daniel Hernandez, one of his professors who had been a visiting professor at UT Austin, recommended that he look into graduate programs there.
Madrid Padilla decided to pursue a PhD in statistics at UT Austin because it opened up the opportunity to work in both industry and academia.
"I'm the oldest son, I have many siblings, and my dad is sick, so I'm like the head of the family," recounted Madrid Padilla. "So I went for something which could make the future easier. I've found that statistics is actually very difficult and interesting."
UT Austin has proved a fertile research ground for Madrid Padilla.
"My time at CIMAT was all about learning technical skills, but UT was quite the opposite," he says. "It has been all about how to have a broad perspective."
Since coming to UT Austin, he has collaborated with scientists from many departments, as well as scientists from the University of California, Davis, Carnegie Mellon University and elsewhere.
In one such collaboration, he's working with a computer science graduate student to improve natural language processing. The goal is to train a computer to effectively understand and predict human language. This means that a computer can complete a task such as completing a sentence. The challenge is that the most accurate methods are very slow at prediction when the computer is trained using lots of data, limiting the possible applications. The UT graduate students hope to improve upon existing methods which are fast, but not very accurate.
After graduation, Madrid Padilla will be going to the University of California at Berkeley as a postdoctoral researcher to pursue research interests in statistics. The focus of his research will be in anomaly detection, high dimensional statistics, and also exploring natural language processing and potential healthcare applications.
Comments