For context, please refer to this thread. TLDR: over the past four years, Georgia Tech has always placed second in our ICPC regional contest, and did not advance to the ICPC World Finals due to the region repeatedly being allocated only one slot.
I would like to try a thought experiment: are there CF users with ratings 2400 or higher, are ICPC eligible for WF20 (based on extrapolations of the criteria for WF19), and would be interested in spending a year at Georgia Tech from mid-August 2019 to May 2020? The tentative plan for funding is at the Masters student rate (full tuition coverage, plus a stipend of about 1500 USD / month), with other responsibilities being TAing for algorithms courses and conducting research.
If you are interested, please email me at [email protected] with your CF id, and ideally, a short CV. Please also note that while I'm extensively involved with algorithmic problem solving competitions, my role with the GT ICPC team is minimal: the main coach/organizer is fahrbach.
Am I misunderstanding the term "thought experiment"? This just straight up seems like an experiment... Or bribery :^)
My guess is that he's trying to see if GA Tech won SER (Southeast USA Regionals), the region would get 2 slots and UCF would still get to go to WF.
Which will probably be the case unless 3 teams from GA Tech beat us, but I really don't see that happening.
This is very much speculative because the final decision in admitting students resides with the administration of my school (with whom I have discussed these ideas before, hence the fairly definite arrangements/workload/finances). I'm trying to do a preliminary call to get an idea of what kind of participation (and projected results) we can hope for.
Re: bribery, please refer to https://codeforces.me/blog/entry/64880?#comment-488143 for a more detailed analysis of the costs. My previous coaching experiences indicate that the (opportunity) cost of training an undergrad student at a school like GT to the CF2400 level is about $10^5: it takes 2-3 years working at 50% time to get good at contests, undergrad tuition is about $5*10^4 per year, and in-contest behaviors are highly unpredictable. So I've always felt that if I do get pressured by my school to improve ICPC performances (aka. goto WF), I should first check whether such resources can instead go to those with far more established interests/experiences.
I don't think it is bribery. Many colleges offer scholarships for athletes, so why not do the same for competitive programming.
Forget even making analogies to athletes. UCF, which coincidentally is the school that beats out GT, literally pays their competitive programmers (quite well, from what I've heard).
Yes, being essentially required to do better than such teams was one of the main reasons that this plan was brought up and investigated. It took a while (3 years to be precise) for my colleagues to start believing my worst-case predictions about WF qualifications though...
On the other hand, good performances in math/algorithms contests can also significantly boost one's chances of being admitted to top US undergrad institutions, whose tuition support (4 * $5*10^4) are an order of magnitude larger. So I believe what UCF is doing is great for their students, and is working extremely well.
Non-resident undergraduate tuition is around $30,000 as listed on this site and this shows that Master's courses have lower but roughly comparable costs; note that non-residents cost more.
Let's ballpark the cost at $30,000 ($1,500; for 10 months plus half of $30,000). That's a lot of money, where would it come from?
Also, looking at typical TA salaries, $1,500 corresponds to 40-hour weeks (taking a typical 21 working days/month) at $8.83/hr minimum wage, or 32-hour weeks at $11 median wage. That's fine. The question is if that's enough for living — university teachers are famously underpaid.
I wouldn't trust that "typical TA salaries" link; salary pages are notoriously inaccurate in general, and being a University TA has a different meaning than just being employed as a "teaching assistant" at, say, a high school. At most universities in the US, if you are a graduate student and a TA for a class, then you will just immediately get tuition covered and be given a stipend. TA's often do most of the work of running and teaching a class (the term "assistant" is loose), so it's really like hiring a lecturer.
That part only matters as much as it affects working hours and I still concluded "that's ok". The real question is whether someone like that would have to get an extra job.
Wow, someone is already doing the math... oh wait, I'm posting on CF.
I'm a bit amazed that you got roughly the right cost estimate (without knowing the backend numbers, and the notion of overhead). The short answer to your question of "where would the money come from" is a mix of school TA funds, and research funds: the latter are in the $4*10^5/3yrs range (e.g. a grant for trying to get faster bipartite matching algorithms on dense graphs). So the costs of such visitors can be mostly absorbed into a 1 year delay in taking on new PhD students.
I've actually never dug up the exact numbers from the front end, but on the backend, the school heavily subsidizes the tuitions of TAs, so the overall effect is much larger than the salary values.
The standard work load for a TA is 10 hours/week. This number can go to 15 in weeks with tests/exams/too_much_due, but never more than that.
Living costs in Atlanta is about median among US cities: rent in house shared with several others is in the $800~$1000 range. So $1,500 is about the minimum needed to live (comfortably): an extra job won't be needed, but also don't expect to save more than a quarter of it.
CodeForcers Assemble!
I've got a few questions which may require you to clarify a bit.
Thanks for your reply in advance. I'm actually quite interested in this opportunity despite the chance seems minimal for me.
Can anybody explain why GT cares this much about ICPC? I've never really understood its importance.
Why do you think so? GT does not pay their competitors. It was the team being unqualified last year that cares so much, not the college..
Isn't GT going to fund these new students?
It is only a thought experiment I guess.. A master student at GT can get the 1500 stipend and tuition coverage as TA anyway. So GT does not offer anything even if this experiment is real.
The `thought experiment' aspect is mostly due to the uncertain visa situations: to my understanding, GT doesn't have as many exchange agreements with other schools, so some of these situations could potentially lead to a lot of paperwork.
I discussed the cost breakdown in https://codeforces.me/blog/entry/64880?#comment-488237. There are still some costs involved, but it ends up being on the low end of research-level resources. A funny aspect about US schools is that faculty usually have their own research funds, so you are correct in that the cost to the school can be minimal at the end of the day, and mostly limited to the resources that they spend procuring visas.
I'm not sure of the exact numbers, but only a small portion of the MS students get TA offers, and even then, it's only a portion of their tuition/stipend (instead of full as you're claiming). You are correct though in saying that this is a much lower level of resource allocation (compared to say, a fully funded undergrad program): this was a major motivation for exploring this option.
The other way to look at it is whether such arrangements, with the potential longer term benefits (e.g. further graduate studies, industry internship/job opportunities), are worthwhile for the high end CF users. This is something that I'm not sure about, which is why I decided to post to figure out.
ICPC related things are the only reason for which I've been summoned by the dean of GT College of Computing during the 4 years I've been here. For reasons that I also don't quite fully understand, administrators care about these things significantly more than research faculty, and my tenure case is coming up next year...
The main `reasonable' reason that I see it is that 4 consecutive years not going, due to losing to a team that has a significantly higher amount of resources, is causing a lost of interest among undergrad students who participate in ICPC related activities.
Maybe it's for reputation/marketing reasons?
The first time I heard of UWaterloo is because it's active on the competitive programming scene (supports IOI teams and has solid ICPC teams). I didn't choose it because of that and competitive programming is not really that relevant to my interests but it helped UWaterloo's visibility in this case.
I think the potential super-regional will definitely create a much more fair environment for the North American teams, and GT will have a large chance to get into the world finals.
I would be highly interested if it were two years ago. The only notable accomplishment during my undergrad is my CF rating and it doesn't seem to help in grad school application, so I would be very happy if my CF rating could help me get a chance to do research and get LoRs in GT with full tuition coverage and stipend. There are probably some other people in the same situations.
+1
you know, solving too many easy solved problems doesnt help solving hard non solved problems.
I 100% agree so I'm not complaining about why they don't care my CF rating.
yeah, that is why people with published papers and a clear research topic are prefered.
Having published papers is not easy for undergrads working on theoretical computer science, and I believe it is not "required" for admission to many top CS PhD programs (except schools like MIT because you have to compete with people like Lijie Chen). I can confirm that most of the first-year TCS students in Cornell didn't have a top/second-tier conference publication before they got admitted.
Without published papers we must convince the admission committees with good SOP and LoRs. I know some people who can show their enthusiasm and potential with SOP and other resources very well: one of my friend got admitted by Harvard, Princeton and some other top schools without publication and LoR from well-known researchers in his area of interest. But if you can't, LoR will be the most important evidence to show that you are actually good. Imagine you are a professor evaluating an application, of course you won't trust a LoR from a person you don't even know. But it is hard for some international students to get in touch with these well-known researchers. That's why I think the chance in this thought experiment will be attractive to these students.
(BTW, for people who are in a similar situation: I heard that CMU theory faculty does care about competitive programming and I did get an interview from CMU in my first year of application. Although I was not admitted at the end, they did mention my ACM-ICPC experience in the interview so I think this is a good enough evidence.)
more than 99% of papers are not real breakthroughs, I think publishing papers is not that hard.
Do you need it to beat top 117 team? Kappa
Oof, unlucky dude.
Yes. And we finally did it (without the experiment)
Welp, I was hoping to stop thinking about this for a few days before revisiting this thread. Guess not, but as this is CF, at least I can provide mostly just the raw info...
So first to follow up about the topic of this thread, I did get contacted, but mostly by people towards end of high school who are applying for undergrad studies. In fact, they all met the requirements for this full scholarship at the University of Waterloo, and were pointed in that direction instead. It turns out this kind of application info for contestants is less well-known than I thought...
Then about this year's WF qualification process. The North American Championship (NAC) did get organized, but ended up making things even worse (yay1, yay2). The initial plan of top 18 teams from it goes to WF got changed at last minute to top team from region goes, then 6-7 slots decided by the NAC. From what I could gather about the regional results, getting second in SER means we need to do better than at least half of Purdue, CMU, Michigan, Toronto, Columbia, NYU, Princeton, Berkeley, Stanford, Duke, UMD, NC State, Alberta, Harvard, Brown, and at least one of UCLA/USC. With last year's team, this is at best a 50/50 chance.
As it turns out, 2 of our PhD students (xyz2606 and akaiNeko) have been to WF before. I in fact spend many happy hours discussing things like dynamic resistance, dynamic connectivity, max-flow (by desert97 and Aaron Sidford, both at Stanford), and min-cost matchings with them (feel free to DM me if you're interested in these topics). So given that our best chance to go to WF is still to win the regional contest, and the hosting of NAC by GT, fahrbach and I decided to form a team (GT5) with these two students and animeshf (who first brought this situation to CF). The actual contest was very messy on our end: if all UCF teams are viewed as one merged team, GT5 was down 1:7 at about 120, but somehow they ran away with it during the last hour. (aside: the set is medium difficulty at best, I can imagine teams of highly ranked CF users AKing it in 120~150 minutes).
So to respond to this post, I feel what we did is still very much along the ideas in this thread, except the admissions came through the PhD program (AlanWaP is in fact also a PhD student here). It's something that I have very mixed feelings about: I was WF-eligible up until the second last year of my PhD studies (at CMU), but always viewed ICPC as more of an undergraduate / student community activity, and perhaps more importantly, have never been in a situation like this before .
I hope a saner NAC selection criteria can be adopted so that we can actually use WF to incentivize our undergraduate students, instead of announcing that we'll travel with two teams to WF, IF we make it. However, as an algorithms researcher, I will continue to apply worst-case analysis in the meantime. So as 2 of our team members are on their last year of WF eligibility this year, the same offer of coming to GT to do contests is still in effect for the 2020-2021 school year.
Congrats!
It has come time to right the wrongs once committed. The below is an accurate representation of 2018 GT.