Update Jan. 28, 2018: Unfortunately, I was not selected to move on to the second round of interviews. Although I avoided any big mistakes in my interview, that was not enough; it seems that I needed a little more "wow." But I'd like to think that I gave them a lot of trouble in the decision process. Regardless, it was an intense and rewarding experience, and I do not regret applying! Being among the elites of the country was a proud moment.
Recently, I was honored to be selected for the first round of interviews for the Hertz Fellowship. I am writing this post at the airport after finishing my interview, in an effort to capture my memories and feelings more freshly. I will update with the result when I hear back at the end of January! Here is an overview of my experience and my thoughts about it.
My interview began this morning at 10:20 AM, when I was shown into conference room C (for some reason it was called "Salon C") in a falsely glamorous hotel near the airport. I walked into the ostentatious red room, covered in gold trim and paisley patterns, and shook hands with my two interviewers. From what I have read on the internet, having two interviewers is somewhat abnormal. But everyone I spoke to was in the same situation.
My interviewers were both kind and unpretentious, despite their distinguished careers. They set the stage by describing the current state of the selection process: they had received about 700 applications, and had invited 103 applicants to the first round of interviews. They intended to offer 40 applicants invitations to the final interview round, from whom they would select 10 fellows. The interviewer articulated that the field was absurdly competitive, and that I should be proud to have made it this far. It was kind of him to open with this - it made me feel much more at ease and ready to get into the interview.
We began by discussing some aspects of the field of robotics. One of my interviewers was clearly a history buff, and kept asking me questions about the history of robotics that I could not answer.
"What is the largest use of robotics in industry?"
"Manufacturing."
"What is the company that uses the most within that industry?"
"I'm not sure, but I'd imagine it's an auto manufacturer... Ford?"
"Nope, GM. How did that work for them?"
"Well, from what I've seen, quite well!" (I've toured a GM engine manufacturing plant that was very successful - it uses a lot of robots)
"Nope, they almost went bankrupt, back around 2000. Why do you think it didn't work out for them?"
It went on like this - it was lighthearded but focused and interesting. He clearly knew a lot about the field, and was quickly finding the extent of my knowledge.
He continued along this line of higher-level robotics questioning, asking me about some of my favorite robotics movies and books, and the flaws and fears they expose. Contained in this discussion was both the high point and the low point of the interview. The low point came first: I mentioned iRobot, and he asked if I had read the book. I said no, and he asked me who wrote it. I knew the answer (I promise) but I hesitated for a few seconds, battling against my poor name recollection skills. It was too late. He jumped in: "Isaac Asimov," and I agreed in embarassment - who can't remember the name of the most renowned sci-fi author of all time, a man whom I respect immensely, and whose essays and novels I have greatly enjoyed? Perhaps not a Hertz Fellow!
I earned some retribution from this by recalling Asimov's three laws of robotics, but was still a little disappointed.
Next came the high point: we transitioned to a discussion of machine learning in robotics, which stemmed from a comment on the realism of HBO's "West World." I recalled a conversation I had recently with a McGill Dynamics and Controls professor, where he spoke of the cycle that seems to be present in the field: in the '70s and early '80s, artificial intelligence and machine learning began to enter the field. However, it was not as effective as an explicit understanding of the governing equations of motion, and soon receded in favor of that explicit understanding. Now, AI and ML are back in force, and this time they are outperforming the controls we can achieve using only these explicit understandings. This rise in learning methods that we cannot fully understand is concerning to many dynamics and controls folks, myself among them.
After explaining all of this, the interviewer said how happy he was to hear me mention it. He added: "Do you know why AI and ML receded that first time around? It was because of me."
He went on to describe how and why he had worked to defeat these approaches to problems in the field. It was clear that I was lucky to have said what I did: I said it to the right man, I guessed "'70s and '80s" correctly, and the interviewer mentioned his agreement with my statement a further two times during the interview.
After this point, the other interviewer took the reins, and we had a discussion about my research experience, what components of it I had most enjoyed, and what parts of it I hoped to pursue in grad school and beyond. I described my most recent research project in detail, and this prompted a discussion about the methods and applications of the research. I do not know if this was a good thing, but this discussion led to a few points when I disagreed with the interviewers, and spoke this disagreement.
This discussion lasted 20-30 minutes before we moved to the only real technical question I received during the interview. The lead-up to the question was interesting: throughout the interview, I said again and again "neural net" and "machine learning." These are both terms that are thrown around a lot, and until recently I had only the vaguest understanding of them. However, I learned in a practice interview with two professors at my university that if you use a term that you do not fully understand, your interviewers will pounce, and question you about the details of that term. They will not indicate whether you are answering correctly until you're far afield, guessing at the finer points of a difficult concept, and it all comes tumbling down. During a second practice interview that two of my friends kindly conducted for me, I found I couldn't avoid using these terms, and they pounced. They asked me what a neural net was, and I stumbled hard.
With this in mind, I had specifically brushed up on these explanations, and a specific example of a neural net, before my interview. With this in my pocket, I unabashedly used the terms throughout the interview, and my interviewers indeed asked: "What is a neural net? Draw one. What does this component do? Why does this work?" They really grilled me, but due to my good friends and my professors, I answered fairly well.
After this, the interview drew to a conclusion. I was invited to ask my interviewers a few questions, and I clumsily asked about the contrast between administrative and feet-on-the-ground research positions, and how to get to a position as a civilian where I could affect the implementation of the technology I develop. They answered patiently and thoroughly, and gave me some great insight and advice.
I believe that the Hertz has the most effective selection process of any of the high-profile graduate fellowships. The NSF, NDSEG, etc., simply cannot match the process of interviews that the Hertz boasts, at its smaller scale. However, there is a lot of room for subjectivity and luck within this tighter selection process (side note: this may not apply to interviews for fellowships, but interviews may not even increase selection effectiveness). The interviews vary widely in content and technical difficulty: my interviewers asked few technical questions, and those questions they did ask were not pre-determined. They came up with the questions on the fly, allowing me to steer the interviewers towards my stronger points, and decreased the likelihood of a question I knew nothing about. They also did not press very hard in the early stages of the interview: they started off with lighter questions, and did not press me on controls during my discussion of my prior research as much as I had expected.
This is in contrast to stories I have heard online, and from other candidates that were interviewing on the same day. Some interviews began with a set of pre-determined technical questions that built on one another, some had only one or two very long and in-depth technical questions, and some had no technical questions at all: just a discussion of the field.
The variations I discussed above make the interview a little bit impossible to prepare for. Nonetheless, the preparation I did was critical.
The most important part of my preparation was practice interviews. I was lucky to have two professors at my school who were knowledgable and accomplished in similar fields to my own, and willing to take an hour of their days to put me through the wringer. From this interview, I learned that I must overcome my nerves at all costs - I cannot fidget in my chair, say "um" or other filler words, or answer unconfidently (even when I am just speculating). Additionally, I must not use terms which I do not fully understand. It is the interviewer's job to pick up when you are using concepts that are at the edge of your knowledge, and ask you to explain them in depth. If you could not explain the concept to an outsider to your field, don't use the term!
Normally, I would not have an issue with this, but the interview is difficult, and more intimidating than I had initially thought. You can be bombarded with question after question that lie just at the edge of your knowledge: you know that the answers are somewhere deep in the folds of your brain, but they are impossible to access. You hear yourself saying "I don't know," "I remember studying that two years ago, but..." "I can't say exactly how to..." and the pressure starts to build inside of you. I found it important to experience panic taking hold as I failed to answer question after question, and to breathe, overcome its grip, and really think about the next one. I got slightly better at either saying "I don't know," simply and strongly when I didn't, or taking the time I needed, formulating an answer, and concisely stating it when I did.
I had two close pals give me a second interview, and they too put me through the wringer. We all got so absorbed in the roles we were playing - them of decorated, lofty interviewers, and me of the lowly undergrad - I was reminded of the Stanford Prison Experiment. But this was exactly what I wanted. Nervous as hell, I was able to practice overcoming these nerves, and answering questions even when the answers didn't immediately surface in my mind.
Based on the weak points I noticed in these interviews, I reviewed the content of my courses that were relevant to my future career: DiffEq, Linear Alg., Probability Theory, Dynamics, Dynamic Systems, Computer Vision, Data Structures, etc. I also reviewed the concepts and terms that were most important in my field. Knowing these terms was the most useful during the interview - it expanded the vocabulary I could use without hesitating, knowing I could back it up with a concise explanation if asked.
The Hertz Foundation's task is monumental: to select 10-15 fellows from an incredibly qualified pool of applicants. The institution rises to this challenge by making an interview process that is unpredictable and difficult, and by this virtue, bullshit-proof. My preparations for the interview - two practice interviews and studying of key concepts from my field - was helpful, but this may have been by luck alone. I approached the interview as a process that would select me if I am qualified for the fellowship, and to eliminate me if I am not. In this way, I was able to step back a little bit: I met with two incredible leaders in engineering whom I deeply admire, and was able to have an engaging conversation with them. I enjoyed the adventure, and the opportunity to practice the high-stakes conversations that I hope I will have in my future career. Whether I move to the next stage or not, the Hertz has been an important and humbling experience that I believe has helped me to grow.