Each week, the statistics department at BYU invites a guest speaker to give a presentation on sometimes statistical. This week's speaker was Dr. Matthew Heaton from the National Center for Atmosphere Research. His talk was titled "Multi-Fidelity, Spatio-temporal Computer Model Calibration using Predictive Processes." In every day lingo? Dr. Heaton gave a presentation directly related to my project.
I'll admit; when I first heard from Dr. Reese (my mentor who got me in touch with Dr. Woods in Southampton) that Dr. Woods' research area was statistical computer design, I not only had no idea what he was talking about, but I also wasn't too excited. I wanted to work with DNA and RNA sequencing, not computer. In fact, I am not the biggest fan of computer programming. I just simply am not extremely talented at it. However, after listening to Dr. Heaton's talk, I understand more of what computer design entails. Moreover, I think it will be an interesting project--one I will not mind becoming my masters' project (see 25 questions).
Computer design is above and beyond experimental design. In some cases, it is impossible or impractical to obtain many data points. Sometimes you only have 10, but the system or model is vastly complicated. Case in point: Dr. Heaton was trying to model the effect solar storms have on the electrical grid. He has data from only one day. This is where the computer comes in. Using various mathematical equations (such as differential equations) and given inputs, the computer model is able to simulate data points to aid in helping predict whatever the end result is. However, this is only part of it. Next comes the stats. Running a complicated program can take an insane amount of time. Dr. Heaton got only 20 data points after running his program on a super computer for one month. Only 20. But, using statistics and mathematical reduction, he was able to compress the dimensionality problem. In one case, Dr. Heaton reduced a 2 billion by 2 billion matrix to a 5 by 5 and 4860 by 4860 matrix. Anyway, without getting bogged down in the muck of mathematics, computer design is not about computer programming; it is about dealing with small data sets and still being able to explain the bigger population. So, I will not be working with thousands of microarrays with DNA. Instead, I get to work with huge, complicated models like the atmosphere and global warming or a rare disease that only affects one in a million.
I realize this learning journal is probably very technical and boring, but by writing down what computer design is has helped me better comprehend the scope of my project.
No comments:
Post a Comment