CS Grad Programs: URSA and Goldwater Scholarships - acmutd/help-guide GitHub Wiki

Research Scholarships and Fellowships

There are a variety of competitive fellowships / scholarships available for you to apply to if you are planning on pursuing a STEM PhD after completing your undergraduate. These typically give a small to large sum of money, but more importantly are extremely prestigious and can open doors for you in graduate admission later down the line.

General Scholarship Resources

UTD Undergraduate Research Scholar Awards (URSA)

UTD URSA is a local UTD scholarship worth a small sum of money ($500). It is essentially a way to begin padding the Awards / Honors section of your CV. There isn't really much to say about it, however I highly encourage you to apply every possible time. It is great practice for developing the writing skills required for the following, more comprehensive applications. In fact, there is a direct line from practicing applying to URSA as a Fr/Sph, then applying to Goldwater as Sph/Ju, then applying to the big daddy fellowships senior year.

Here are my two successful URSA applications: 2018, 2019 (I didn't receive it freshman year). It also gives you the chance to practice delivering your research to an audience of laymen on presentation day.

Goldwater

The Barry Goldwater Scholarship is the most competitive undergraduate scholarship you can apply for. Each undergraduate institution may only select four students TOTAL (and recently also one transfer student) to apply for the scholarship. They are usually in their 2nd or 3rd year (seniors cannot apply). The award is a modest $6k, but the scholarship itself is extremely prestigious, and if you are awarded it you should expect to go to a top university. For transparency, I was nominated by UTD but unfortunately did not receive the award.

Given the competitive nature of the scholarship, getting nominated by UTD can prove to be quite difficult. However, usually two out of the four UTD nominees are able to receive the scholarship each year. So you have a decent chance if you can get nominated. Nonetheless, note that there a bunch of factors which we probably aren't aware of, e.g. field specific selection. For example, far more students get the Goldwater in Biology than Computer Science: a quick count yielded around 16 CS recipients vs. ~70+ biology recipients in 2020. However, don't let this discourage you: the Goldwater is great practice for the much more valuable NSF GRFP (which potentially funds your graduate studies, see below).

If you are interested in applying for the scholarship, I would first read as much as you can about it, then contact Dr. Douglas Dow from the honors college ASAP. He is in charge of the entire process and if you are nominated, he will grill you and your application like Korean BBQ to make it as strong as possible.

Getting Nominated

This was the least transparent part of the process for me since I had not applied to any competitive scholarships before. The internal deadline for UTD students is separate from the national Goldwater deadline, since UTD needs some time to select its four students. (Note: Although UTD may select five students if one of them is a transfer student, unless you are a transfer student you area really only competing for one of the four spots). Typically, you need the following to get selected:

  • Near perfect GPA
  • Extensive research experience
  • Hopefully some publications
  • 3 Letters of reference

Here is my entire (and final) Goldwater submission. This was after I was informed of my nomination by UTD (e.g. it was my second submission), so it reflects what UTD and Dr. Dow think is an "okay" submission to the national Goldwater competition. The initial version you submission may still undergo significant changes once you are nominated, since Dr. Dow will grill you on each of the questions and your motivations.

This can be difficult (especially the last two) if you are only a sophomore. However, you can apply both your sophomore and junior year, and I encourage you to do both. Even if you are not selected your sophomore year, you'll have a much clearer idea of the process and perhaps stand a better chance the following year. Nonetheless, even getting nominated is quite good, you can position yourself near the top of all STEM students at UTD (who are planning to pursue a PhD).

The application is generally organized into two sections: a couple different personal statements and also a research statement. After you are nominated, Dr. Dow will help you significantly edit and improve your application. I will share my final submissions for each of these two parts, but be aware that yours will probably improve drastically with the help of Dr. Dow. Finally, your research advisor(s) are extremely helpful in this process. The writing style here is nearly identical to how they write research grants to try to get funding for their own research endeavors. Make use of their expertise and try to get them to read drafts of your work! Especially in the process of getting nominated, they are your best source of information.

I am also always happy to review your application for the internal deadline (to get nominated)! I was lucky to have a previous Goldwater Honorable mention review mine, and without her help I doubt I could have gotten nominated. Just shoot me an email with the info listed on my personal site.

Personal Statement(s)

When you sit down to start writing all the parts of the Goldwater application, keep in mind that this is a competitive research scholarship. (Nearly) everything in the application packet should be a demonstration of your research skill, experience, and potential. One exception is the "diversity" prompt, where you can share experiences that impacted your educational history. Nonetheless, I am sure the most successful applicants have research itself as the core topic in every part of the application. Let's try to break down some prompts from (my year's) application.

* In one or two sentences, describe your career goals and professional aspirations (see example below). 
This statement will be used in publications if you are selected as a scholar or honorable mention. 
(Example: Ph.D. in Molecular Biology. Conduct research in biomedical science and teach at the university level.) 
(200 character limit including spaces.)
Response

Ph.D. in Computer Science. Conduct fundamental research in theoretical machine learning investigating generalization bounds of neural networks while teaching at the university level.

Notes

In 200 characters, you want to try to share the "theme" of the research you are keen on conducting. Try to read through what previous scholars have said by looking through them here

* What are your career goals and professional aspirations? Indicate which area(s) of mathematics, science 
or engineering you are considering pursuing in your research career and specify how your current 
academic program and your overall educational plans will assist you in achieving your career goals 
and professional aspirations. This is an expansion of the brief description of the career goals 
and professional aspirations you provided above(3000 character limit including spaces).
Response

Neural networks (NNs) are at the forefront of applied machine learning (ML) due to their wide applicability and powerful representational power. However, theoretical ML has not yet provided a satisfactory description of their performance on data not used to train them—also referred to as their “generalization”—which restricts the use of NNs in sensitive fields. My career goal is to apply computational learning theory (CoLT) and optimization techniques to better define and understand the generalization of NNs. I will lead a team of researchers to describe new theoretical tools for providing strong generalization bounds for NNs while simultaneously teaching at the university level.

Optimization algorithms used to train NNs—such as stochastic gradient descent (SGD)—have been shown to learn a restricted class of solutions. During my Ph.D., I plan to jointly study CoLT and optimization to expand upon NN generalization bounds for more powerful optimization algorithms than SGD. I also plan to utilize CoLT to prove bounds for particular classes of NNs, such as weight agnostic or binary NNs. Current CoLT techniques offer only vacuous bounds for generalization in regular NNs, and my work will investigate whether it is the models or tools that may need improvement.

To prepare for the theoretical work during my Ph.D., I am enrolled in the computing scholars honors program, which readies a small group of CS students for research careers through fast-paced and theoretical courses. I am also pursuing a degree in pure mathematics and am taking several upper division classes that will serve as the foundation for my work within CoLT and optimization. Furthermore, I am currently enrolled in a graduate course on optimization in ML and plan to take another in convex optimization. Finally, I participated in ML research last summer, and hope to participate in an optimization research experience this summer. My rigorous course load as a double major will prepare me for the foundational theoretical work within ML during and following my Ph.D.

At the UTD Advanced Networks Research Lab, I mentor a group of undergraduate student researchers by explaining relevant concepts in ML and CS theory. These experiences, combined with teaching during my Ph.D., will strengthen my ability to instruct at the university level.

Existing CoLT techniques designed for general classifiers may be antiquated for NNs. As a professional investigator, I will build on the experiences gained with CoLT and optimization techniques to develop new tools suited for analyzing NNs. I plan to create novel optimization algorithms for NNs which work closely with CoLT for stronger bounds, and describe additional restricted classes of NNs designed with bounds in mind. I believe that building new algorithms and NNs designed from the ground up for generalization bounds will yield practical algorithms with better theoretical performance.

Notes

This is the first meaty response. It goes without saying that for each of these responses you should draft, draft, and then draft again, all while having friends and professors giving you feedback at each step. My approach to this prompt was:

  1. First introduce a problem (or in particular, a set of problems) in the (sub)field I am interested in researching, why they are important, and state my goals.
  2. Next, I propose a class of solutions which I am interested in studying.
  3. I then state why I am / will be qualified to conduct this research.
  4. I close with a summary and plan to achieve my goals.

This is the general format for scholarships (e.g. GRFP personal statement). The lengths of the different parts may vary. Although there may be some leeway in following this format, it is always a sure place to begin your drafts with. Just make sure you are answering every part of the prompt!. You would be surprised how many responses fail to do this!

* Describe an activity or experience that has been important in helping shape or reinforce your 
desire to pursue a research career in science, mathematics or engineering. (1500 character 
limit including spaces)
Response

My goal of pursuing CS research was actually discovered in an orthogonal field: astronomy. At the start of high school, I wanted to discover planets outside our solar system and investigate mysterious astronomical phenomena throughout space. The summer between my second and third year of high school, I reached out to a local astronomy professor working with data from the McDonald observatory and explained my motivation and goals. Fortunately, he offered me the opportunity to help him with his work on discovering exoplanets, enabling me to find out what astronomy research actually entails.

I worked almost exclusively on the computer, coding simple Python programs to analyze the vast library of available data. Under the guidance of my professor, I transformed data and ran existing algorithms to search for the small decreases in brightness caused by exoplanets passing between individual stars and Earth. Looking back, it was not the astronomy part, which was both unfruitful and not the romantic endeavor I had envisioned, that made the summer exciting. Once I was able to write functional code with the help of my mentor, I was elated at the fact that I could search through thousands of data points at the touch of a button. I was also intrigued that my simple programs could detect minute outliers invisible to human experts. Later in high school, I built on my experiences in data analysis and began investigating more complex pattern recognition tools, ultimately leading me to ML.

Notes

Looking back on this response, this is the "cheesiest" part of my application.

* Goldwater Scholars will be representative of the diverse economic, ethnic and occupational 
backgrounds of families in the United States. Describe any social and/or economic impacts you 
have encountered that influenced your education - either positively or negatively - and how 
you have dealt with them. (1500 character limit including spaces)
Response

My Nepali grandma was a math teacher. When visiting Nepal, it was a tradition for my mother and me to attend her class. Since math has little language barrier, I could work on problems with my grandma’s students. This group problem solving offered a unique contrast to the memorization-based math education back home. After my first visit to Nepal in elementary school, I began enjoying math more. I started solving problems in my free time and later even competed in regional math competitions. I visited Nepal every two years and would be excited not only to see my grandma and her students but also to vex them with new problems I had encountered.

As I grew older, however, I became conscious of the difference in education and facilities when compared to my school in Austin. Although I felt geographically powerless, I eventually realized that I did not need to go to Nepal to see differences in educational opportunities. One example was a local Austin district which lacked sufficient resources to promote STEM. While I could not directly improve conditions in Nepal, I did have the ability to change things locally. Along this vein, in high school and university, my peers and I have organized free math and CS programs at low-income middle schools.

My experiences in my grandma’s classes ultimately inspired me to study math in university. I am grateful that my education was impacted by my grandma’s students, and I hope to one day eventually return to Nepal to teach a math class for her.

Notes

This was by far the hardest portion of the application for me, as I really didn't have any obstacles from this category. I had to dig deep to find something relevant, and even that was not relevant to the research I wanted to eventually conduct. My advice is to be honest here (especially if you are a minority or come from a tough background: own it!), but if you can tie the prompt back to your research I think you'll be even better off! Research dense is the way to go!

That's the bulk of the application. There are other parts where you describe research projects: my advice for those would be to describe exactly what you contributed to the project. E.g.:

Description of research, including your involvement in AND contribution to the project.
(1000 character limit including spaces)
Response

Maximum margin classifiers are a family of classifiers where the decision boundary they learn sits exactly in between the different classes of data (if the data is linearly separable), thereby maximizing the margin, or distance, to each class. Our project attempted to create maximum margin neural networks in order to improve the performance of neural networks (NNs) on unseen data points.

I implemented our method for margin maximization in Tensorflow, testing and debugging it on a variety of binary image datasets. Our method sampled random points from hyperspheres of varying radius centered at each real data point. I utilized open source image processing packages to load in images, then sampled from hyperspheres centered at each image in data space. During training, this “fake” data artificially pushes the margins of our learned NNs towards max margin boundaries. This work was an independent study between a faculty mentor and me.

Notes

etc. etc. That's pretty much it. There is a "research skills" section, but I don't really know how much that matters...

Research Statement

This is one of the most important parts of the application, and is what you should spend a good chunk of time on. Again, here is my entire application, and the research statement is the three pages at the very end. My advice: sit down with your faculty advisor early to come up with this project. Then, write many many drafts. You can also use a project which you have already begun (like I did). This is where your experience applying to URSA will be helpful. Write many many drafts and ask your friends to look it all over each time. Your faculty advisor(s) also have plenty of grant writing experience, and can help you out a lot if you pester them enough. Good luck!