Finding Your Thesis Statement & Story

After a few years in a PhD program, working on whatever research project your advisor had funding for, or whatever project seemed interesting or fun at the time, it can be daunting to find a unified story or thesis statement that accurately, concisely, and interestingly summarizes your work. In this post, I will describe how I was able to find a cohesive story that unified the disparate strands of work I’d completed, and helped me focus my thesis proposal into an interesting final project.

Background— My Work So Far

My thesis work began in reinforcement learning, using differentiable decision trees to try to make RL agents more sample-efficient by leveraging human knowledge for warm-starts. Then I developed a method that discretized those differentiable decision trees into discrete trees, making interpretable models that humans could inspect. After that, I took a hard-pivot into language-guided reinforcement learning, and influence-functions as a tool to explain properties of word-embedding models (long story for how those projects came to be). Then, on an internship, I did some really cool and interesting work on personalized federated learning (which I was very proud of and wanted to include), and finally I had recently completed a large-scale user study on human perceptions of explainability. So this was a lot of topics and papers across an absolute smorgasbord of subjects.

Step 1: Group Your Projects

The first and easiest piece of advice I have to offer is to group your projects into meaningful areas. Ignore the in-depth mechanics of your projects and just think about the high level problems, insights, or contributions. Are there natural clusters that emerge? Can you find common themes across some of the papers? Again, you don’t need to find a silver bullet for all of your work, just try to reduce the number of subjects that you need to unify.

Step 2: Summarize Your Work

For me, this meant talking to as many people as possible about my work. I would give people a summary of everything I had done, and eventually I got sick of saying the same things over and over. I found ways to take shortcuts or group projects, and eventually found a useful few-sentence summary of my work that conveyed what I most cared about in my completed projects. For me, this was:

“I’ve done a bit of human-in-the-loop RL, trying to integrate human expertise into RL agents to improve their generalization or their sample efficiency, I’ve done some personalization in my time at Apple, and I’ve worked on human-centric interpretability/explainability, both on contributing new methods and on understanding how humans use those methods. For my proposal, I’m planning to bring those all together by learning to personalize explanations of a task-centric agent alongside a human in a user-study.”

Step 3: Re-Examine Your Work

After you have this summary that ties your work into a clean narrative, you may need to revisit your completed projects and re-frame them in this new light. Originally, my differentiable decision tree project was more about sample-efficiency than about the human in the loop. While I kept almost all of my original results, diagrams, experiments, and slides, my script changed to focus more on how this was an interactive learning approach that leveraged human expertise and put the human front-and-center. While this wasn’t exactly the original pitch, it allowed me to present a cleaner and more cohesive overview of my work that wasn’t as jarring for my audience.

As part of this step, also consider revisiting your proposed project! If you find a clean story that unifies your work, it may also reveal a natural capstone project to bring everything together. Follow that thread and consider how you might complete the narrative with your final project. If you see a natural end-product, your committee likely will do, and they’ll be expecting you to go there.

Step 4: Find a Thesis Statement

This is maybe the hardest part, but, as I’ve mentioned previously, it doesn’t need to be a single sentence. So don’t stress too hard. My thesis statement was three parts, one for each body/cluster of work that I had completed. My thesis statement slide read:

Interactive and explainable machine learning yields improved experiences for human users of intelligent systems.

1. Machine learning with human expertise improves task performance as measured by success rates and reward.

2. Personalized machine learning improves task performance for large heterogeneous populations of users.

3. Machine learning with explainability improves human perceptions of intelligent agents and enhances user compliance with agent suggestions.

One overarching “vision” statement, and 3 testable statements that I would prove in my talk.

Conclusion

Finding a story for your work is hard and does not come naturally or easily to many. Give yourself plenty of time, talk to people both in and out of your field, and try to view your work from a new angle. It’s almost always possible to find a satisfying narrative, it just takes work, creativity, and re-framing to see your projects in the right light. Good luck!

Preparing Your Thesis Proposal Presentation

Your oral defense (your thesis proposal presentation) is where you get to convince your committee (and a wider, public audience) that you deserve a PhD. You need to show that you’ve done cool work, you’re a smart and capable researcher, and you have thought of an interesting final project that will bring home a strong body of work. No pressure!

In general, a thesis proposal presentation will look something like this:

1. Introduction

2. Problem setup/motivation

3. Related work (for the entire thesis)

4. Thesis Statement

5. Completed Projects (tie back to thesis statement regularly)

6. Interesting open question (setup for proposed work)

7. Proposed work (final project design and expected findings)

8. Timeline

Introduction & Motivation

As stated above, your introduction should set the stage for your problem. You don’t need to dive right into the technical details of your work— start by getting your audience to pay attention. You should open with a brief introduction of yourself, and then get into the motivation for why your work is important or interesting. What problem out in the world are you interested in? What gap in the literature needs to be addressed? What unique combination of fields or ideas have you happened upon? Find your interesting hook and motivation, and use that to drive your presentation. Ideally this will somehow tie into everything you have done, so you can hark back to it throughout your presentation.

Related Work

The related work section is tricky, because many times your thesis projects can draw from different areas and have drastically different bodies of related work. My advice is to pick the few papers or projects that will be directly referenced in this talk, and possibly some work from your committee members as well (assuming they work in your area, which they probably should). I put ~30 papers up on screen for my related-works slide, broken up into 4 sections (1 for each “section” of my talk). This included my papers, in bold, so that my committee/audience knew that I had made contributions to each section. Of these 30 papers, I only went into detail on the 3 or 4 that were most directly related to my talk, then moved on.

Thesis Statement

Committees will vary on how seriously they take your thesis statement, but you should take the time to get it right either way! Your thesis statement does not need to be a single sentence, so don’t stress finding the silver-bullet sentence that perfectly summarizes your work. But do make an effort to present a testable hypothesis (or a couple) as your thesis statement, and have it be something that you do test in your research. If you were to show your thesis statement slide in a vacuum to an audience in your area, they should have a good sense for what to expect (and hopefully, be interested in it!). I’ve written a whole separate post on finding my thesis statement and advice for finding yours, if you’re interested!

The Work You’ve Done

This bit is the easiest! You’ve already completed several projects and, ideally, given talks at conferences or seminars on your work. Take the talks that you have already done, and staple them together! Now, as mentioned previously, you may need to view your projects in a new light in order to get them to fit into one cohesive thesis, so take care to actually rehearse the presentation and ensure that it makes sense. But overall, this part of the talk is largely pre-made, all you need to do is make tweaks and fixes to get everything to flow nicely.

The Work You Will Do

The last couple of sections of your talk should set up and then describe your actual proposed work. What gap has been left in the papers that you just talked about? What interesting project would unify the various strands of work you’ve presented or completed? How are you going to complete that project, or answer that question? You should present this as one of your completed projects, but in future tense. So talk about the methods that you will employ, the data analysis you will do, the results you expect to find, and the implications of those results for the field at large.

Timeline

The timeline is a required part of your thesis proposal at Georgia Tech, so don’t forget it! In my proposal and in most that I’ve seen, it’s one simple slide with a Gantt chart or spreadsheet on it, where you will essentially say how long you expect everything to take. You might say something like “January-February: Design model, March: Run study, April: Analyze results, May: Write & Defend.” It honestly doesn’t need to be much more in-depth than that.

Parting Words

Once you have your presentation, I cannot emphasize enough how much I believe you should run it by your committee early. Please do this, it is so helpful. Once everything is set to go, all that you have left to do is propose!

Responding When You Don't Know Something!

Acknowledge When You Don’t Know Something

A big part of being a good researcher is recognizing when you don’t know the answer to something, and a likewise a big part of your PhD milestones (quals, proposal, and defense) will involve testing this. I’ve spoken with several professors and other students about this, and it is considered a universally important subject to these PhD milestones— your committee will test whether or not you know what you don’t know, and have the strength to admit that.

Areas You May Be Pressed

Your committee will likely try to back you into a corner where you don’t know the answer to something, usually in an area that you should be the right person to answer the question. Whether it is some obscure model or technique you haven’t heard of, some hypothetical about how your work would apply to new datasets or domains, or questions about fundamentals in math that you haven’t brushed up on, your committee will likely search for (and find) things that you don’t know. And they will know that you don’t know. If there is one thing professors are good at, it’s sniffing out bull-shitting students.

Responding If You Don’t Know Something

It is okay to not know everything and the right thing to do is to admit and acknowledge it. You can (and maybe should) speculate about what might happen if your model were applied to a new dataset, but you should make it clear that this is speculation and you would need to do the actual research to know for sure. If you really haven’t heard of something you feel like they expect you to know, you can respectfully apologize for not having heard of it, and ask for more information on where you can learn about it. If you are being asked to compare your work to something else you haven’t heard of or read, offer to brush up on the other papers and then follow-up offline. Or again, speculate on how the related work might compare, based on what you know about how it works vs. how your technique works. But if you haven’t run the experiments, do not make claims that you cannot support with evidence.

Summary

Most committees try to get students to admit that they don’t know something. Failure to do so will reflect poorly on you. You shouldn’t proactively try to not know something, or fail to prepare adequately because you can just be honest and say you don’t know. But at the same time, don’t back down or be afraid if you don’t know the answer immediately. Your committee expects this, and they want to make sure that you know it and can handle it professionally. Be confident, be factual and correct, and you’ll do great!

The Thesis Proposal Process at Georgia Tech

I recently completed my thesis proposal at Georgia Tech, and the process included several questions that I did not feel were adequately documented anywhere. With this post, I’m hoping to provide a straightforward timeline/guide for the thesis proposal process at Tech for future students. Admittedly, this post ended up much longer than I expected it to be, so maybe treat it more as a reference material and come back to each section as you get to that stage.

Who is this for?

Ideally this would be useful for any PhD student at Georgia Tech, but I am in the Computer Science PhD in the school of Interactive Computing. I believe the process is slightly different for different departments/PhDs, and I will make a note where I know things are different. I don’t know the full process for Machine Learning or Robotics PhDs, but the overall guidelines here should apply.

The Core Timeline

Below is a rough timeline of what your proposal process will look like. Some steps (like 2-5) can be jumbled in order, and more details on all of these steps are below. I would suggest allowing for ~4-5 months for this process. Scheduling professors can be very difficult and slow.

1. Identify your committee and reach out to ask them to join you in this lengthy process.

2. Schedule your proposal (see below)

3. Write your proposal document and send it to your committee for approval

4. Prepare your proposal presentation

5. (Optional) Present an abridged proposal to your committee members one-on-on and get their feedback in private.

6. Announce your proposal to the school (must be 1 week before the actual proposal, see below)

7. Propose and get your forms signed!

I’ve written separate posts for each step of the process, all linked above. For more information about each step and for a LaTeX template with some of my thesis proposal, refer to those posts to learn more. Below, I’ll briefly cover scheduling, because it isn’t interesting enough for its own post.

Scheduling Your Proposal

I’ll be honest, this is just very difficult and I don’t have a lot of great advice for you here. I used whenisgood to find good times, first sending it to my advisor and getting his availability, then removing times that he couldn’t be there and sending the modified version to my committee members. I would recommend confining your times to 9-6 PM, and provide weeks of available days. Literally weeks. Professors have negative time, and finding a slot where 5 of them are free for 2 hours can be maddening. More than anything else, this is why you need to start so early. So do not put it off.

Don’t forget, once you have your date, you must send out an announcement to the school one week before you propose. This is as simple as finding one of the dozens of existing announcements in your inbox, copying the format, swapping in your information, and sending it onto the PhD Program Coordinator. Include a title, date, time, room number or virtual room link, and an abstract for your talk.

Choosing Your Thesis Committee

Your committee ultimately decides whether or not you pass your proposal and your defense, so you want to choose them carefully!

The rules from the CS Handbook state that:

“Your advisor is the chair… Your committee should be formed of at least five people. At least one faculty member must be from outside CoC (from another unit at Tech or an outsider, and should have a Ph.D.), but the majority must have some CoC faculty appointment.”

Note: You only need to have three people on your committee at the time of your proposal, but it is recommended that you have all five members at that time.

So you only need 3 members, and at least 2 must be from the College of Computing (CoC) and 1 outside of the CoC for your proposal. However, it is strongly recommended to have your entire committee together (so that is 5 members with at least 3 from the CoC).

I recommend working with your advisor and other students in your lab to identify good CoC faculty for your committee— they will know which professors align best with your research direction, which are the most helpful on committees, and what those professors look for in thesis proposals or defenses. All of this is stuff you should know about your committee before you propose!

Choosing an External Committee Member

Fortunately, choosing an external committee member is quite straightforward. When I talked to the graduate coordinator for my proposal, I was told that the only requirement was that my advisor agree to my external committee member!

The external committee member can be a great networking opportunity, and many people suggest reaching out to other big-name or rising-star professors in your field from other universities or institutions. I personally went with an advisor from another research org that I knew I worked very well with, and who I wanted on my committee for his useful feedback and direction.

Reaching Out to Your Committee

This can be slightly scary and intimidating, but it is part of a professor’s job to be on thesis committees, and I am given to believe that many of them do enjoy it to some extent. Be respectful and concise and provide as much useful information as possible, and send weekly reminders if you don’t hear back! My thesis request emails looked something like:

Subject: Thesis Committee Invitation/Request

Dear Professor <Name>,

I am in the process of writing and preparing my thesis proposal on interactivity and explainability in human-centric machine learning, and I would greatly appreciate the opportunity to have you on my thesis committee.

My thesis will encompass a selection of works that have already been published on interactivity and explainability for machine learning (including X, Y, and Z), works that are currently under review for personalization and human perceptions of explainability, and a proposed final project on personalization and explainability in the context of a user study.

I plan to get a written thesis proposal to my committee in early March, and to do the formal proposal presentation in mid-April.

Please let me know if you are willing to serve as a committee member for my thesis proposal!

Thank you,

And these were sent sometime in mid-January. So, about 2 months before I planned to finish my written proposal document and about 3 months before I planned to do my proposal. For more info on the overall timeline and other details of the proposal, check out my other posts on the subject!

Writing Your Proposal Document

When to Write Your Proposal Document

I suggest starting as early as possible! As soon as you know you are planning to propose soon, start the document. It will take a couple of weeks to finish, maybe more, and your committee should ideally have ample time to read and review it before your proposal. As indicated above, I gave my committee one month to review my document before my proposal. They do not need to officially tell you they approve or anything like that before you do the proposal presentation, but when they approve your thesis proposal and admission to candidacy, they are approving your document. So they need to have time to read it.

How to Write Your Proposal Document

Warning to other departments: Robotics has a different written format requirement! I don’t know about ML, but in this section I will again be speaking only to CS.

Again, the CS Handbook states that:

The PhD. proposal consists of:

1. A proposal document written by the student in which he/she will propose the research work to be undertaken as part of his/her doctoral work and a schedule for completion.

2. An oral defense of the proposal document

Helpful. So there’s no information there on format, length requirements, background to the proposed work, etc. In practice, you are free to interpret this how you will, so long as you clearly write down what you propose to do, why you are proposing to do it, what you expect to see, and you provide a timeline for completion.

For my proposal document, I went ahead and wrote version 0 of my entire thesis (click here for a .zip with stripped down source files to get the template that I used, as well as a completed intro chapter just as an example for your reference). This included all of my relevant prior work, introduction and related works chapters, everything that would be part of my final thesis. The final document was over 200 pages, so it was a herculean undertaking for my thesis committee to read, but I actually do recommend doing it this way, for a couple of reasons.

1. Your committee sees the entire trajectory of your PhD and gets a better sense for your expertise and what to ask you questions about.

2. You get free feedback on your thesis before your final defense (one day).

3. I had at least 1 committee member tell me that they really liked seeing the full document, and that they wished more students did it this way.

For more information on the thesis proposal timeline and other aspects of the thesis proposal process, see my other posts on the subject!

Get Early Feedback On Your Thesis!

This is an optional extra in your thesis proposal and defense process, but I cannot recommend it enough!

Why Get Early Feedback?

Once you’ve written your document and nearly finished preparing your presentation, I strongly recommend bringing it to each of your committee members individually and getting their feedback. This has several benefits, it gives you nice one-on-one facetime with your committee members and gives you a chance to get to know them a bit better, it helps them to see your story early and therefore sets their expectations for your proposal, and it gives them a chance to give you honest feedback in private. Rather than airing their grievances in public before an onlooking crowd, you can have a back-and-forth discussion in private.

How Should You Prepare?

I suggest taking your proposal presentation and slashing it down to the barest of bones. My proposal presentation was 90 slides, and my “condensed” presentation was about 20. I included a high-level motivation (2 slides), a thesis statement (1 slide), and then for each project that I was going to present (step 5 above), I included a “methods” and “results” slide (2 x 6 papers = 12 slides). Then I included 3-4 slides on my proposed work, and my timeline. I stepped through each of these with my committee members in one-on-one meetings, giving them a speedy-version of what I was going to say in my proper, full presentation.

How Early Should You Do This?

I recommend doing this at least few weeks before your public presentation. One or two of my committee members originally misinterpreted what I was going to do with my proposed work, so this extra time gave me a chance to re-calibrate their expectations in a one-on-one, rather than in front of everybody. I got to rework my thesis statement with the help of other professors, I got early feedback on my proposed study design that I took back and integrated before even presenting the proposal publicly. I also got useful feedback that some results were missing or didn’t make sense, so I knew which completed projects I had to provide more or less information on in order to clarify my story. In the closed-door session of my proposal, I had more than one committee member say that they didn’t have any questions for me because we had already been over everything in a one-on-one. Seriously, I cannot recommend this enough.

For more information on the thesis proposal process, see my other posts on the subject!

Delivering Your Proposal Presentation

This post assumes you have already prepared your presentation, written your thesis proposal, and ideally gotten feedback on your proposal!

Once you’ve gotten this far, you’re nearly done! By now, you have written your complete document and your committee has had a chance to read it, you’ve prepared your presentation and script (hopefully with feedback from your committee members), and you have scheduled this date when all of your committee members can be present at once! The process of the proposal is determined by your chair (your advisor), but in general you will be introduced, you’ll present your work, you’ll answer public questions from the committee and from the audience, and then the audience will be kicked out and you’ll answer private committee questions. Then you will be kicked out, the committee will deliberate, and you’ll be invited back in to hear the results. Congratulations (hopefully)!

The only major piece of advice I have to offer during your proposal presentation (apart from the usual: practice, be confident, know your stuff, and relax!) is to recognize and respond appropriate if you don’t know something. I’ve written a separate post on this (linked), but it is something you should expect your committee to search for, and you should be prepared to respond professionally (and to generally remain calm, it’s okay to not know something).

After you finish delivering your presentation, all that’s left is to get everything documented!

Getting Your Forms Signed!

Don’t forget to get everything signed so that you officially enter candidacy! Look on the Georgia Tech website for the Request for Admission to Ph.D. Candidacy form, and send it to the right people on DocuSign. You may need to talk to the grad coordinator to figure out who the right people are for the admin signatures, because unfortunately those people are constantly shifting in the CoC.

Thesis Proposal Summary

The thesis proposal is a relatively long process! If you do it the way that I did, you come out with a nearly completed thesis, plenty of useful feedback to carry over into your proposal project, and an admission to candidacy! I think the most stressful part of all of this is (1) the scheduling and (2) finding the right story for your work. Most of us don’t have a clear thesis road planned out when we start, and we have to sort of reverse-engineer a story around the projects that we’ve completed. To recap, the loose timeline I suggest includes:

1. Pick your committee. Ideally make it your final defense committee, and make it people who will support you and who will provide you with good advice and guidance.

2. Write your document and your presentation. Give yourself plenty of time for this.

3. At the same time, schedule with your committee to find a date & time for your proposal.

4. Send your document to your committee and set up 1-1 meetings for a whirlwind preview of your thesis proposal presentation.

5. Send out your 1 week notice

6. Propose and get your forms signed!

Good luck!