How I Fail: Hal Daumé III (PhD’06, Computer Science)

For this post of How I Fail I’m interviewing Hal Daumé IIIwho is an Associate Professor in the Computer Science department and Language Science Center at the University of Maryland. He has an awesome website and is on Twitter @haldaume3.

1. Thanks for joining the How I Fail series! Please introduce yourself. What are you doing now (position, area of expertise, location), what did you before to get here? Do you already have any “failure statistics” you would like to share?

I am currently an Associate Professor in the Computer Science department and Language Science Center at the University of Maryland. I started here as an Assistant Prof back in 2010, after having been an Assistant Prof in the School of Computing at the University of Utah for four years. Before that I was a grad student (at the University of Southern California, working at the Information Sciences Institute from 2001 to 2006), with a stint as a a summer intern at Microsoft Research in Seattle in 2003. Before that, I was an undergrad at Carnegie Mellon in math (1998-2001), and spend the last half of my undergrad time working at the Language Technologies Institute. Before that I was in high school!

All of that sounds pretty good, though basically every transition had a number of “failures” that were notable at least to me :). Going chronologically….

  • When I applied to undergraduate programs, I really wanted to go to Stanford. Many of my friends would be going there, it would keep in in California (I grew up in LA), and I just really liked the place. I didn’t get in. My second choice was UCSD; I also didn’t get in there. This was very clearly my failure: I just wasn’t a very good student in high school. The fact that I got into CMU was, to me, a minor miracle, and it was by far the best college I was admitted to (I still sometimes wonder how that happened).
  • As an undergrad, during the summer before my last year, I really wanted to participate in the Johns Hopkins summer workshop, to which I applied and was not accepted.
  • When I applied to graduate school, I applied to precisely three places: CMU computer science, USC computer science and Stanford linguistics. I had had a really difficult time deciding whether to apply to Stanford linguistics or Stanford CS, and in the end I made the decision to apply to linguistics for a stupid reason: Stanford CS quals terrified me. I did not get in to Stanford (probably largely because I had very little linguistics at the time, and my statement was likely garbage). I was really upset about this at the time, but in retrospect I’m actually pretty glad they rejected me. If I had gotten in, I would have gone, and I’m pretty sure I’m happier having gone to a CS program than a linguistics program.
  • I’m not sure if this counts as “notable” but it was a big deal for me. I submitted my first real paper in grad school to ACL 2001 on “document compression.” I wrote the first draft on my own before giving it to my advisor (Daniel Marcu). I considered myself a pretty good writer: it was something I had always enjoyed and I took some creative writing courses in college (I originally wanted to double major in creative writing). A day or two later, Daniel gave me back a marked up copy of my draft; one would have had a hard time finding a square centimeter on the page that did not contain red ink. I was so upset, I left the office, plopped down on my bed and cried a bit.
  • I’ve had plenty of papers (and now grant proposals) rejected; for papers, I don’t really remember which ones, with a few exceptions. In 2003, I started getting really interested in machine learning, and I started trying to learn as much as I could. (All of my previous work was in natural language processing.) This was relatively soon after the original latent Dirichlet allocation paper, but before “topic models” were really a “thing.” I had developed a very of LDA that generalized to n-grams rather than unigrams, and I wrote it up and submitted it to NIPS. Neither I nor any of my close colleagues had every submitted to NIPS, so I didn’t have much guidance here. The paper didn’t only get rejected, but got solidly walloped by the reviewers.
  • The other notable paper rejection for me was the paper that made up the majority of my dissertation: the paper on the “Searn” algorithm, which eventually was published in the Machine Learning Journal. This was probably the most frustrating publication experience I have ever had. I can’t even remember in what order we submitted it, but we submitted it to (at least) two NIPS, one ICML and one AIStats starting in 2006, and it was roundly rejected from all of them. Eventually we gave up and submitted to MLJ, where it got in after (I think) one round of revisions in 2009.
  • I applied for both faculty positions and industry research positions when I was finishing my Ph.D. in 2006. The biggest disappointment for me at the time was getting an interview at CMU but not getting an offer. The second biggest was not getting an interview at Microsoft Research in Seattle.

2. Do you keep track of your failures (rejected papers, grants, job applications…)? Why/why not?

Only in my head :). Until recently, I’d never really thought of keeping an explicit list of failures. I still elect not to do it. All of the things I listed in the previous section were really hard for me to deal with at the time; I take rejection to heart, deeply. This has always been something that I’ve struggled with as an academic: positive reward is pretty rare (and often very delayed), and negative reward is pretty frequent. Yet, I am really happy with where I’ve gotten: I’m at a really great university with a great group of colleagues (yay CLIP!), I live in DC which I love, and I have a ton of flexibility in what I work on, how I work and when I work (and when I don’t!). I don’t plan on starting to keep track of failures, explicitly: I guess I prefer to focus on the positives.

3. What do you think about sharing failures online? Are there disadvantages for researchers who do it?

I have mixed feelings about it. I think if everyone did it, that would be one thing and could be useful. But in most cases, when I see a CV of failures, it can read a bit like a humble brag. Probably the list I provided reads this way in parts. If someone has produced a CV of failures that I actually look at, that person has probably “made it” in some sense, and so a CV of failures can often look kind of like “despite all this adversity, I still managed to be successful!”

I think it’s worthwhile that students and junior researchers know that everyone gets rejected and everyone struggles and everyone wants to quit at some point. Because if we all can acknowledge that openly, and get rid of as much posturing as possible, I think that opens the door to providing better support for our colleagues and friends, and making the research community a more welcoming space in which I, at least, would be happier to belong. I don’t currently plan on keeping a CV of failures, though honestly I’ll be really interested to read more arguments in favor in other entries in this series: maybe I’ll change my mind.

4. What do you do when you receive a rejection? Do you have some process/ritual of dealing with failure? Has this process changed throughout your career?

My general strategy is: avoidance :). When I get a paper or grant rejected, I just don’t read the reviews for at least a few days, maybe a few weeks in the case of grants (and course evaluations). Of course eventually I do, and in the beginning I’d just get really sad and defensive. Over time I’ve developed the following strategy: I try to think honestly and seriously about on which points I agree with the reviewers and on which points I disagree. On the points where I disagree, I try to put into words why I disagree. And then I go for a hike or a walk or go to yoga or something mind-clearing.

5. What about when you receive good news? Who do you share the news with, do you have some rewards for yourself?

Mostly I share the news with my partner, unless it’s something significant enough to be considered “news worthy” (like a funded project or award or something) in which case I share with the CLIP lab and the UMD PR team. The nice thing about sharing with the lab is that you then typically get back about 25 congratulations emails :). Sometimes celebration might include wine, though naturally after going home.

6. If you would have a CV of failures, do you think it would show more/same/less failures than other people in your field? What factors (strengths/weaknesses, circumstances of your job, prestige of lab/university…) do you think influence this?

I imagine the specifics would differ, but the scope would be similar. The biggest differences I can imagine would be based more on what people choose to go after rather than what they’re rejected for. For instance, I’ve never had a paper rejected from Nature, but that’s because it wouldn’t really occur to me to submit a paper there. In terms of factors that correlate with failures, I’d expect them to be basically the same as factors that correlate with successes, and that you’d see a “rich get richer” effect in general, and would probably see similar correlations with demographics as in other areas of academia.

7. Can you share some examples of failures which hurt the most, and why that was?

I already kind of answered this above, but probably not getting an offer at CMU was the most difficult thing for me. At the time, I personally cared a lot about recognition and prestige, and if I had gotten an offer I would have accepted almost immediately (despite the fact that when I was an undergrad at CMU I was rather unhappy). Add to that the fact that I knew, from working at LTI as an undergrad, all of the people involved, the rejection felt far more personal than rejections from other places. (I know now that I shouldn’t have felt that way, but that didn’t help me 10 years ago.) More generally that job search season was difficult for me: I ended up with three offers that I was really happy with, but I went into the hiring seasons a bit arrogant and when I didn’t get as many interviews, or offers, as I was hoping for.

8. Are there any opportunities you didn’t take that you wish you had (so you could add them to either your “success” or your “failure” CV)?

Not really, though there’s probably choice-supportive bias going on there. The closest was when I applied for the position at UMD that I have now. UMD was hiring “off cycle” and so when I applied here, it was the only place I applied. In retrospect, I probably should have applied more broadly. There’s a good chance I would have chosen to come here anyway (in terms of a balance between being a great place to do NLP, and a city that I love living in, there are very few places that I would rank higher).

9. Can you think of something you accomplished that felt like a success, but you wouldn’t normally add to a CV?

I don’t put “declined” things (i.e., I was offered something but I turned it down) on my CV, and there are a few of those, which I suppose would count. I’m also not going to list them here—sorry! More broadly, the things that come to mind here are personal anecdotes that come ever so often in the form of emails or stories from former students (grad or undergrad) that say nice things, or when I recently found out that someone in our community learned about NLP from my blog while in high school. That’s the sort of positive, long-term reward that’s great, but doesn’t really make up a CV item.

10. Is there something we can all do to improve how failure affects others in academia?

I think we can try our best to de-personalize things. I talked about this on my blog a while ago, but I try to avoid using the word “you” in reviews. I try to talk about the work or the paper or the presentation or the research, not about the authors thereof. I think we as a community should also acknowledge that academia—like basically everything else in life—is not a meritocracy, and to recognize than when we as individuals fail, it’s not necessarily a reflection on our value as researchers. And I think that we need to have institutionalized mechanisms for “on boarding” new members of the community, so as to minimize as much as possible an in-group and out-group mentality.

10. What is the best piece of advice you could give to your past self?

There’s more to life than your publication and citation count.

More How I Fail?

Thanks again for joining the How I Fail series! If you – the readers — have any suggestions regarding this series, please leave a comment below. You can also join the weekly newsletter if you want to make sure you won’t miss any posts!

1 thought on “How I Fail: Hal Daumé III (PhD’06, Computer Science)”

  1. Pingback: How I Fail: Michael Ekstrand (PhD’14, Computer Science) – Veronika Cheplygina

Leave a Reply