ERSP Application

  1. What do you hope to get out of participating in the CS Early Research Scholars Program?
  1. Explain how your status as a minority in CS has affected your experience in CS so far OR Explain your understanding of how being from a underserved/minority group affects a person’s experience in CS and how you came about this understanding.
  1. How would your participation in the CS ERSP program help increase the diversity of students in the CS community?

Having been in the AI club all throughout high school and being the president in the last two years, I have spent a lot of time learning the ins and out of machine learning. I'm aware that there exists a large bias problem in the constructing from AI; everywhere from how and where the data was collected to preconceptions of those who label the data. As they say, garbage in, garbage out. I've seen it first hand when working on AI side-projects. When I found out about the William Wang's work in fairness in NLP from last year's ERSP cohort, I felt that this was the perfect opportunity to work on such critical problems in AI that I've experienced so vividly first-hand.

In my junior year of high school, I, along with a bunch of friends, set up a hackathon initially with the primary purpose to serve students in my high school. However, after a lot of work in fundraising, we realized we had enough funds to go beyond just a few students coding for 24 hours in the library. While talking to other schools in the area to drum up interest, we got the same questions, such as whether we could provide computers or whether we could provide a free ride to the venue, over and over. Often, we would get emails back from schools saying that they'd be incredibly grateful to offer their students such an opportunity but were concerned whether there would be enough interest among students, given that there was never had any computer science programs in the school and that many students didn't even have access to their own laptop they could bring to such an event.

We quickly realized that, to no one's surprise, many of these concerns were coming from schools of underserved neighborhoods of the county. After getting connected with a few students at those schools and talking with them, it appeared that many of them were hard-working, motivated individuals, but just felt that the field of CS was just beyond their reach; we found out it wasn't a lack of interest in CS, but that many lacked the confidence that they had the brains to be able to break into such a field, even as we could tell that many had more than enough potential to do well in CS and then some.

This stigma of "not being smart enough" seems to have to do with how many of these students had no access to resources, classes, and extracurriculars that many of those in my community had from a young age to develop knowledge in CS. After seeing how well-executed the hackathon projects or initiatives done by the more privileged students of the area, perhaps they felt that they just simply couldn't compare in terms of talent or intelligence, when really it was that they just did not have the resources to practice and develop the skill over many years like many of those in my community.

It was an epiphany when I realized that there were so much emphasis and organizations designed to facilitate students into STEM in my community that it became clear that many students going to these clubs or after school activities were doing it simply because their parents forced them or to increase chances at getting into a prestigious university. It seemed that supply of CS resources outpaced actual demand. In many underserved communities, it was exactly the opposite problem, where genuine passion for CS and STEM was outpacing the supply of resources.

With this realization that there was so much untapped potential in genuine interest in CS in such communities, we changed objectives; our goal became to become the spark to as many students as possible to break into CS. We moved our venue to be closer to underserved communities; we guaranteed computers for anyone who could not bring their own; we provided workshops by knowledgeable individuals from the industry and students geared to teach complete beginners in technologies such as HTML, CSS, and JavaScript or Python and backend web development; we made sure to place a lot of emphasis that total beginners were welcome in both our marketing material and website; we set our target attendee demographics to be at least 50% students of underserved backgrounds.

In the end, it was a great realization to me of how much potential underserved students had; it made me so happy to have helped guide the first steps of many underserved, passionate students in their adventures in CS, just as this event served as the first steps to fix this supply-demand gap of underserved communities and CS resources.

The research I hope to work on is to develop methods for bias correction in machine learning. I feel that, more and more, AI is used in impactful decisions--everywhere from deciding the optimal punishments in judicial cases to facial detection systems to catch criminals. However, many of such systems, due to the biases of the data (many datasets are simply lacking in one or more racial or underserved groups) or how the focus of research primarily benefits one group of individuals over other (e.g., the focus of European languages in NLP over Afro-Asiatic ones).

Additionally, in many of my projects, such as Recommeddit, an NLP recommendation algorithm based on the consensus of recommendations on Reddit, I find myself needing to read a lot of research papers to build on existing knowledge. However, a lot of these papers assumed knowledge of many areas of NLP that I wasn't knowledgable of, and I found myself spending a ton of time researching and just trying to understand these papers when really, it would've been optimal to get a quick understanding of them and get straight to building the projects. I feel that ERSP will give me the foundations that I need to quickly parse and grasp previous research to do my own research, build some projects that make use of such previous work, and just for fun to stay up to date with the latest evolutions of the field of AI.

Furthermore, doing a master's program in AI has been something that really appeals me; with research being the ones to develop cutting-edge models and industry implementing and optimizing them for scalability, to me, I tend to think that the former would be something that would be more interesting to me. Having worked in a lot of practical AI projects, anywhere from hackathon projects, such as Archiscape, which would build a 3d model of a house by simply taking pictures around it, to months-long side-projects, such as Azar, a time-estimation tool based on past data on how long your tasks took in reality compared to how long they actually took, as well as taking multiple internships where AI was applicable, such as at UnMesh, where we scraped recipe websites and used machine learning and regexes to convert the recipes into structured data, I feel like I've gotten a good glimpse of what practical AI looks like in industry. However, I have yet to get my hands on any AI research, so I believe ERSP will get me my first experiences of what it's like to work as an AI researcher, which will give me a good way to compare between research and industry and figure out which one would be right for me moving forward.

As something that is largely disregarded by industry but nevertheless an important part of developing robust AI systems, I feel like my aim in research is

While my demographics are well-represented in this area of study, I hope my hyperfocused and attention to detail thinking can provide a different kind of thinking when it comes to this project. I look forward to working with a group of undergrads who can bring all sorts of perspectives to the table.

When you think of someone with ADHD, you might imagine that they would have trouble staying on one task without getting distracted, but for me, I have the exact opposite: hyperfocus. Once I find something that interests me, I can't stop myself from diving deep into everything there is to the particular subject; hours past by without me noticing, and I can get sucked into such tangents even to the detriment of my academics.

Over the years, I've developed strategies to mitigate some of the most harmful effects of hyperfocusing (such as writing schedules and writing down distractions for later research), while also using it to my advantage. As someone with a genuine passion for AI, I let my curiosity go wild as I went down innumerable rabbit holes and have come out with a lot more knowledge than many of my peers in the subject.

In grade school, my teacher had let me know of how good I was at drawing; "that owl he drew," she said, "had such detailed strokes and even the eyebrows were shaded so accurately." "Ohhh, so that's what it was," my parents would reply, laughing. I don't blame them: seeing as only the top left of the owl's head had been finished, it would be a bit difficult to know what I was drawing!

Another side effect of my hyperfocus is my extreme knack and attention to detail. As someone who does web development, I could have fun for hours, tweaking the layout and aesthetic of a UI to be just right, where others might be OK with an 80% polished design. While this attention to detail could sometimes lead to spending too much time on details that were not that important, I have found workarounds to such problems (such as breaking down tasks into goals every 2-3 days to keep pace), and, just as with my hyperfocus, taken advantage of its upsides to paying attention and pointing out small but important parts of a project to teammates that may have otherwise gone unnoticed.