Summer internship at CMU

Hello World!

Update

So I have been pretty absent on my blog for a few months. This is beacause while working with Professor Sankaranarayanan at Carnegie Mellon Image Science Labs and applying to PhD programs, it's been a busy past few months and I haven't had time for much! So for the next few days I will be updating and talking about various parts of my research and applications.

Carnegie Mellon Image Science Labs

Working at Carnegie Mellon Image Science labs has been among the most interesting experiances of my life. For one, the culture was vastly different than the culture of the labs at Davis, the main difference being the size. Most of the labs at Davis are fairly large, and I was typically assigned to work with a graduate student rather than working directly with the professor. At CMU, I would meet with Professor Sankaranarayanan once a week and if I finished me week's assignment early, I could meet with him any day I wanted to review my work! This was not something I had every experianced with any of my other labs, where I would have to wait for the professors and sometimes they would even miss meetings (missing meetings isn't too unusual for professors who are very busy).

My project for the summer was to design an autoencoder which analysed the primary components of hyperspectral images. A hyperspectral image is an image which contains very precise data regarding the distribution of the different wavelengths in each pixel. An autoencoder is a system that uses convolutional neural networks to compress and decompress the image.

autoencoder

This autoencoder functions a bit differently however - when decompressing the image, it uses linear algebra and properties of hyperspectral images rather than machine learning. This way, the compressed image contains useful infornmation. In this case I designed it so that it predicts the primary spectra of the hyperspectral image, thus indicating what sort of materials are inside the hyperspectral images and what reactions might be taking place when the image was taken. This could be useful for chemists, biologists, and climate scientists.

Submission to Journals

After doing this work, I had to do the extremely important next step of submitting it to conferences so I could get credit for it. I submitted it to conferences at Rice University and Purdue University where I could give poster presentations. Unfortunately these were virtual. But at the Purdue conference, I was actually able to meet Professor Beecham who asked to talk with me afterwards and recomended that I apply to Purdue to work at his lab!

One of the cool things about going to conferences is that sometimes, randomly good things happen. I met one of Professor Sankaranarayanan's graduate students at the Grace Hopper conference and that is how I got the internship. I also got this connection at the Purdue Conference. But there are many times, such as the Rice conference, where nothing really comes out of it. It's kind of hit and miss, but the hits are really great!

In addition to poster conferences, I also submitted to 'real' conferences where they actually publish your abstract and everything. I got rejected from both of them at first. However I also applied to the student undergraduate consortium for the AAAI conference, which is supposed to be harder to get into than the regular student abstract submissions. However, I was rejected from the student abstract track and accepted to the undergraduate consortium! Or rather, I was waitlisted and then eventually accepted. This meanas that my abstract will be published in the conference proceedings!

However, there was a problem. Four of my PhD programs had their applications due on December 1st but I wasn't taken off the waitlist until the second. That meant that four graduate programs didn't know that I was sole author on a paper accepted to a major AI conference! Thankfully, two programs easily let me revise my application. Caltech at first did not, but I emailed somebody else and they updated my application. UCLA however, did not let me update! But then I realized something: I had said I was first author on the abstract in my application, because I thought that I would be first author. But in reality, I was sole author because of some confusion on the manuscript submission. I told UCLA that I actually had the authorship wrong, which is a really big deal. So they agreed to change it!

Anyways you can see my manuscript on my site, and soon I'll have a three minute lighning talk uploaded because AAAI said I need to make one because they are mean lightning talks are helpful in explaining research.

May you be ever victorious in your endeavors.
M.E.W