Metis Project 3

I hope not to be this exhausted again for a long time.

Today I got Metis’ Project 3 done. Last Friday, May 1 my third (and probably final) application for a Research Associateship through the National Academy of Sciences / National Research Council was due. These two things really did not belong in the same week. I’m not even that short of sleep… ok, I’m pretty short of sleep… but I’m mentally exhausted simply from the intensity of what I’ve been working on. I am so looking forward to recovering some condition and placing my focus solely on the bootcamp.

Igneous Rock Classification

The thing about research proposals… well there are several things. The one I have in mind is that they’re talking about doing something without actually doing it. So it was a relief to go back to my Metis project and do what I could with the remaining time. I’ll have to write more about it later when I have more energy and focus, and in any case I definitely want to go back and extend this work a bit for my portfolio on Github.

I wanted to give myself a break for this project and work on really interesting data I already understood, and that led me to EarthChem. The goal of this project was 1) to perform and present interesting work using classification algorithms and 2) build and show some skill with either SQL, cloud computing via Amazon Web Services or Google’s competitor offering, or advanced visualization. In pursuit of 1), I ended up working on the question of using overall rock chemistry to classify rocks by geologic eon (Archean, Proterozoic, or Phanerozoic). The Earth has changed over time… cooled off, for certain… so how much has this changed magma and rock chemistry? If you throw enough chemical features in, can you always distinguish an Archean rock?

If you ask the question globally, the answer I got in a limited amount of exploration is “no.” That said, with random forest or gradient boosting forest techniques, I found that with the addition of sample location data (lat/long), it becomes possible to predict with some precision and even good recall. Then, at least in the test region of Southern Africa, if I focused down to create a specific regional model, the results can be quite good. Is this just because there are only a few broad rock types in each eon in that region, or most regions? Probably. It could still be an interesting tool for geologists, whether from a pure research perspective or from the standpoint of some application like metal prospecting. Check out my slide deck and, for the really curious, my Jupyter Notebooks for the full journey of exploration.

RAP

What am I up to with the RAP fellowship? I’ve been struggling for my whole career caught in a dilemma between a lack of confidence in myself, a compulsive tendency to try to intuit what I’m supposed to be doing from the people around me, and the knowledge that I am sitting on a massive amount of talent in terms of mathematical and reasoning ability and really want to use it for something. I want to put that talent to use, but have found it very difficult to take the necessary steps. I didn’t grow up with that kind of role model, haven’t really been able to connect with such people so far in life, and have kept changing strategies. I started three or four separate research careers, depending on how you count them, then drifting into teaching, and then drifted out of that. I needed to take that time to finally work on a ton of things about myself, but here I am at 41, and what am I going to do with the rest of my life?

I just revised the About page today, where I note that I want to be at the junction of chemistry, crystal symmetry, quantum physics, and electomagnetics. I knew this 20 years ago, so it’s definitely time to take action. Last summer I was exploring a lot of options, networking and applying for positions, and that eventually led me to the RAP program, and once inside that system I found out that one large group of people with interesting projects were working in solid state physics at the Air Force Research Lab outside Dayton, OH. That’s a 2 hour drive from where I grew up. Who knew? I won’t name names at this point… hopefully this version of the application is finally a success, and I can talk more then… but I wrote to a bunch of people on the list of sponsors, got a few positive responses, and at this point have spent at least 2 months’ full time equivalent work on a research proposal for deep bandgap semiconductor defects for quantum devices, including a machine learning portion related to optimizing the parameters for synthesizing defects.

–PAG

Blogging platform assembled by Jekyll, Poole, and Zach Miller of Metis.