Getting a Computer Science PhD in the USA
parentheticallyspeaking.org
parentheticallyspeaking.org
Well that's a lie. NC State is $20k for a CS PhD Student, oh, and that's before you pay the $4k a year on student fees.
The article glosses over it, but you're paid dirt. And what sucks more, is you're paid good dirt, so all the other PhD departments look at your with their barely 10k stipends and think you're ungrateful for your poverty line wage.
Two examples of many:
- https://ethz.ch/en/the-eth-zurich/working-teaching-and-resea...
There are also students from the French/Italian part of Switzerland, so there's a natural convergence towards English rather quick, at least that was my experience.
(take this with a grain of salt, it's been a while and my information might be outdated)
Depends on the country and programme, but more and more so. Even France, traditionally not at all known for programmes at any level fully in English, has been making huge advancements in those areas, and there are full programmes in the Paris Saclay university hub entirely in English, exclusively to attract foreigners who don't speak French.
https://www.cis.mpg.de/cs-max-planck/
"During the preparatory phase, students take courses to build on their existing knowledge of core computer science areas. Depending on a student’s prior academic background (e.g., MS degree), course requirements may be partially or completely waived at the time of admission."
or here:
https://ellis.eu/faqs#can-i-apply-if-i-don-t-have-a-msc-degr...
"Rules in Europe vary. Some participating universities require a MSc, others accept BSc students as well (sometimes under specific conditions). You will find this information in the portal if you click on the list of this year's recruiting advisors."
In some places you technically don't even need any degree, but in practice you wouldn't get accepted without one. (I was looking into this as an option for someone recently!)
Also notice that in Europe you can get a bachelor and masters in 4-5 years (3+1 or 2), which can be the length of a US undergrad degree.
Source: currently applying to grad school
Better source: https://www.cmu.edu/stugov/gsa/Campus-Advocacy/Stipend-Repor...
Um, ok, ouch... jk :D
There are ways to increase your pay and doing summer work is one of those. However, the student fees portion of my comment still rings true. You'll be spending about $4k on student fees, $500 on parking. Depending on your martial status, number of kids, type of apartment, you will still be barely scraping by.
Source: All but dissertation
This post is based on a Zoom call he conducted a month or so ago for folks and I signed up and attended.
Otherwise, I totally agree with you. Stipends are very little money, especially if you’re supporting others.
Further, in my experience, few professors actually do any research nor write papers. The vast majority of this work is done by PhD students and to a lesser extent post docs. Thus, PhD students are the linchpin that holds the entire "do research to publish papers to write grants to do more research" cycle together. If all PhD went on strike, the research output of a university would dry to a trickle, which would make obtaining further grants difficult to impossible, thereby depriving the university of that luscious 50% overhead.
Said another way, PhD student are net value producers for the university. Forcing PhD students to enroll in "thesis hours" that cost 45k per year is simply a cleverly worded scam to pilfer even more off the grants.
1) you don’t have to pay FICA taxes, which is a big deal.
2) you don’t have to pay tuition or taxes on the remission. That can be almost $30k per year before you reach candidacy.
Even if you love research, the programs will select for people who have good grades/institutions, who lack student debt, can invest 4 years at 20k/yr and survive, and are willing to turn down 6 figure jobs in industry.
Given this, it’s not a surprise that 79% of CS phd students in the US are international, where undergrad costs are often substantially lower.
http://nfap.com/wp-content/uploads/2017/10/The-Importance-of...
For example, when I was a high school student applying to Caltech's undergraduate program I took a look at the graduate school section of the catalog just to see what their math and physics PhD requirements were.
They had a language requirement for a math PhD. To get a PhD you had to either be fluent in one of French, German, or Russian, or you had to have a sufficient reading ability in two of those to read the current and historical math literature in those languages.
That kind of thing is much better to take care of in high school or as an undergraduate.
This a thing of the past, from a time when people in the USSR still published things of note in Russian (the same for the other countries/languages your mentioned). Today everyone publishes in English and I don't of a single math department that still has that as a requirement.
I also don't know of any other such extracurricular requirements for any program (can't even imagine what it would be - community service???)
Go to language exam.
>Students may request to substitute the Italian language exam if it is relevant to their area of mathematics
Which is far far short of "learn a language". But these requirements were always like this (a pretentious formality completely insignificant in comparison to the actual research and course work).
Is it even possible to get a medical or law degree for less than 100k these days without rare scholarships and grants? Even a shorter 12 quarter program still runs around $11k a quarter. And even then the school can’t be bothered to pay for basic supplies and equipment you’re required to get, nor for the thousand-dollar licensing exams. And they still expect to get paid for that one quarter you go offsite to do unpaid work for some other provider to get real-world experience.
What a racket. (It’s probably obviously I’m peeved about a specific case :p)
But law? That should be pretty cheap to provide. It doesn't change overly much and it is just books. You could probably video entire lecture series and just replay them... It is not like they need some sort of lab work with expensive equipment.
Trouble is, there has been an oversupply of lawyers. Salary distribution is bimodal, biglaw pays entirely different than smalltown law, and with a degree from a random backwater it's guaranteed that you'll never make it into biglaw. Those plans would have been an invitation for students to get into debt that they'd never be able to repay. (We are seeing a re-run, this time with pharmacy schools.)
Medicine is a cost center for universities, so medical degrees are expensive. But in a functioning society you need physicians, and no medical graduate with a choice will settle in rural Mississippi.
What I'm saying here is that education as a market is not a good model to have.
It's actually just a little under the median household income for the area, so theoretically it's doable even with kids, though that speaks more to the level of poverty in Philadelphia.
In any case, whatever the downsides, I’d wager that the majority of people don’t know that you get paid to get phds. I don’t think that is misleading.
Ideally you're working on your own research project, not someone else's.
I've been in the industry for several years and made enough savings to cover my current lifestyle for a long time while also paying a significant stipent to a university.
Would it make sense to consider good PhDs where I wouldn't be able to get a salary, or are they always 100% scams?
- subject matches my interest,
- we seem to get along.
Okay, I'll probably also insist on a few practical things (sufficient time to keep some pace, awareness of what conference attendance costs, satisfies the institution's phd programme entry requirements, etc).
But other than that: sure!
In 3.3, you mention “ The value of a standardized test is that it give sus some calibration.” Which appears to have a misplaced “s” after give.
Lastly, thank you for writing such a detailed post!
Hypothetical question for US-based folks: I wonder if there is a non-traditional path for people who can afford to research without the need for a stipend?
Technologists tend to be in the upper income brackets. For a lucky few, it's possible to achieve financial independence in their 40's. At that point conducting research may seem like an attractive intellectual pastime.
These hypothetical PhD students can sustain themselves for 4-6 years without a salary. I suppose there would still be some grant-seeking, depending on the scale of the research. Technically they could actually pay the university, becoming a funding source instead of a burden...
If someone had $2-$3 million in their retirement accounts, it seems strange to scratch and claw for $300k of research stipend funding.
Has anyone heard of such an arrangement? Is this a thing?
So yeah, I don't see why you couldn't do this. You would just set up a private research grant with a faculty member and fund yourself with it.
Gee, I got a Ph.D. in applied math; it could have been called computer science. But I never went through even 10% of that analysis. I didn't get accepted to Stanford, Berkeley, CMU, or MIT, but I didn't apply to any of them. But I did get accepted to Indiana University, Cornell, Princeton, Brown, and Johns Hopkins and went to one of those.
Why did I get accepted? In my application I showed that I had (a) a good ugrad pure math education, (b) a good career in applied math and computing with some good work results (samples), and (c) a lot of learning of applied math and computing since my ugrad education.
While I was a grad student, I regarded the courses as only vague suggestions. If I found the courses not well done or not much relevant to my interests, then I felt free just to ignore the courses. In the end, I had four courses that were good, one of which I already knew maybe 80% of the material (it totally pissed off the prof that I already knew so much of the material and effortlessly and unintentionally led the class on all the grading by wide margins). The other three courses were good and I learned some good stuff well.
Next were the Ph.D. qualifying exams: There were five exams, and I led the class on four of them. For three of the four, I led because of what I knew before I entered the program.
Next, what really got me on track, waxed the bottoms of my skis, to get my Ph.D. was that in one of the good courses I saw a problem, saw no solution in the literature, and in two weeks found a solid solution. Later I published the work. Any of the faculty members would have been proud to have published that work.
For the dissertation itself, I identified the problem and found an intuitive solution before entering grad school. Next, independently in my first summer, I converted the intuitive solution to solid math. Later in a few weeks I wrote and ran software that did well illustrating that my solution worked, was practical.
So, summary lesson: (i) Arrive well prepared, from ugrad school, work experience, independent study, whatever. (ii) As soon as possible, hopefully before entering grad school, independently get a research problem and make good progress on it. (iii) While in grad school, likely largely independently, hopefully find some little problem, find a solution, and publish that. (iv) Likely independently, finish the dissertation project, submit the written document, and stand for an oral exam. (v) Do expect never to pay any tuition, but don't expect any pay. Instead, have some money, a car, clothes, etc. from, say, savings from work before grad school. (vi) Don't expect much from the department -- don't expect good research projects, good hints for project solutions, etc. Instead, just work and produce good results almost completely independently.
One more piece of advice: For a Ph.D. in any of the STEM fields, start with a good background in pure and applied math, say, foundations, abstract algebra, linear algebra, and analysis. In particular, learn how to write math. Then outside of pure math departments, will likely be the local star in abilities in math. Then for research, mathematize some part of the STEM field.
Instead of "the norm", I took the subject to be essentially how to get a US computer science Ph.D. and gave some ideas that worked for me and are "not the norm".
Three points that might help:
(1) My suggestions about doing so much independently can cut out a lot of slogging through dangerous waters of organizational behavior, competitions, comparisons with other students, personal relationships, department gossip, personal image, etc.
I've seen a lot of people seriously hurt in graduate school. In the program I was in, I thought that the students were beautifully qualified, but only about 1 out of 16 left with a Ph.D. In two words, it was a blood bath. From the failure rate, that program, and others I've seen, made the Army Rangers and Navy SEALS training programs look like fuzzy-bunny play time. Nearly all my fellow students were perfectly capable of doing good research quite independently; their problem was interactions with faculty and the department on the way to doing good research. Soooo, it can be helpful, if a student can, to work so independently, if only just to reduce possibly harmful interactions with faculty.
"Seriously hurt": I've seen very capable, dedicated people get seriously hurt, sometimes for life, and sometimes DIE, as in DEAD, from the damage. One who was hurt was my brother. One that was badly hurt was a physics prof I knew. I saw a lot of profs that had been badly hurt and developed really sick personalities. One student who was hurt had been Valedictorian, Summa Cum Laude, winner of Woodrow Wilson and NSF fellowships, astoundingly capable, totally dedicated, and DEAD.
Let me put it to you this way: Sometimes a university will have as their official requirement for a Ph.D. dissertation "an original contribution to knowledge worthy of publication". Soooo, two options: First option, do the work, write it up, print it out, drop it on the desk of the advisor, say you believe that the work is publishable and you are ready to stand for an oral exam. Second option, publish the work and drop the dissertation on the desk of the advisor with a copy of the letter from the journal accepting the paper.
(2) At least at one time the Web site of the math department at Princeton stated (from memory) "No courses are offered for preparation for the qualifying exams. Graduate courses are introductions to research by experts in their fields. Students are expect to prepare for the qualifying exams on their own. Students should have some research project underway in their first year.". Okay, that's essentially what I did and got it done before I saw Princeton's statement. So, what I did was essentially "the norm" at the Princeton math department.
(3) Fields of research vary. There is math, computer science, the rest of the STEM fields, the social sciences, the humanities, the arts, biological/medical research, and more. But here we are talking about computer science and, appropriately enough, I've tacked on some math. So, in such work it is common for a new Ph.D. to get an assistant professor slot where they are expected to publish, hopefully get research grants and publish, get promoted, have successful graduate students, and get promoted, etc. In all of this, they are wanted to do good research, good enough to win prizes, get grants, give students a good start on their careers. The "good research" is to be at least "new, correct, and significant". In nearly all of this work, the prof has above him to look up to for guidance only God; otherwise the prof is working essentially "independently". Uh, a suggestion then, along the lines of what Princeton said and what I did, is to learn to work heavily independently and to learn that as soon as possible. Right, it's not "the norm". In the program I was in, "the norm" was 15 students out of 16 leaving the university without a Ph.D. No doubt, of those 15 students, interactions with faculty were not always sweetness, smiles, and good coffee and Vienna desserts.
For your "rambling", that sounds like I was being graded in an English class! One reason I got into math was that my correct proofs could not be criticized as "rambling" or anything else by any of the teachers. Instead of "rambling", I just described step by step what some of the issues were and how I was successful. What I wrote is MUCH shorter than the OP -- one of my points was, and I started with it, the OP was too long and complicated. If you want to apply "rambling", do that to the OP -- it covered lots of stuff that is just too far from what is crucial: What is crucial is one word, research. Two words, publishable research. Four words, prize winning publishable research.
Ah, computer science has long regarded a good algorithm as one with worst case performance O(x^n) for any problem size x and some positive integer n. Okay, this definition was from Jack Edmonds. He left the University of Maryland (UM) without his Ph.D. Then he published some of his work on networks. Then some of the faculty at UM came to see Jack and assured him that should he wish to stack some of his papers and put a staple in one corner, that would be accepted for a Ph.D. Sooooo, Jack did his work independently.
For the "two weeks", that's just what it took. I had and applied good preparation for the problem. I found the problem in a relatively advanced and well done course -- soooo, it's not too surprising that the problem was sitting there unsolved. It was an old loose end in the subject, and I tied it off. Some of what I did was a bit surprising. But when get out to the edge of a subject, there can be some problems still left and relatively easy to solve -- it can help to have a good math background, and that other people didn't have such a good background is some of why the problem was still unsolved. When get out to the edge like that, not all the remaining problems are ready for a Clay, Turing, Abel, Nobel prize -- there can be some little problems also.
My POINT was not that I solved a problem in two weeks. Instead, my POINT was, as part of how to get a Ph.D., if a grad student can find some such doable problem with a publishable solution, then finding that solution can "wax their skis", cut out department gossip, cut down any issues of image or presentation of self before the public, let the student "prove themselves", etc. Or, to heck with the research result; instead, do such a thing to solve essentially POLITICAL problems. Soooo, yes, part of getting a Ph.D. is to do well enough on politics.
Bottom line: If a student can work relatively independently as I described, then that can ease getting a Ph.D. Right, it's "not the norm".
Oh, CMU was mentioned: One of my (official) dissertation advisors was later President at CMU! He seemed to like my work -- he had some of his other students build on my dissertation!
Ah, one more piece of advice: Where to find problems such as I did for my Ph.D.? Sure, the usual way is to get a problem from an advisor. Hmm .... Here is another way, not "the norm": Get a problem off campus, from the real world. E.g., I don't know but I can guess that in the server farms of some major corporations, big banks, Google, Facebook, Amazon, Microsoft, Cloudflare, the US DoD, etc. they have some unsolved practical problems, e.g., how to do well on performance, reliability, security, scalability, and cost all at the same time. So, there, pick a problem, confirm that the literature has no practical solution, build a suitable mathematical model, find an optimal solution, write some software to confirm that the solution is practical, and publish that. So, for the criteria "new, correct, and significant", the "significant" part can come from the fact that some 10s of millions of dollars are to be saved, or reliability is greatly improved, or .... For "new", you checked for that at the beginning, and besides such server farms have been growing so fast that it is unlikely that they have good, old solutions for all their problems. Also, maybe go for a future problem assuming some factors of 10 scale up; by the time you get your work done, that future may be the present! For "correct", that can be mostly just about the math, and there can establish correctness with mathematical proofs. In short, that is what my Ph.D. dissertation did -- so, it can work. Also, surprising or not, commonly profs do NOT really have feet on the ground, finger tip feel, good knowledge of all challenging practical computing situations. So, should be able to find a good problem to solve.
Is this advice "great"? Well, it was "great" for me -- it worked, and I came out undamaged. Also it's not "the norm" which might be an advantage!
Re length/rambling: It's not a personal attack on you, just a suggestion that the information you write can probably be compressed and made much more effective. Yes, this is not an English class. And yes, you're trained as a mathematician and trying to be rigorous. But often times dropping some details to convey the general ideas is sufficient, especially on a fairly technical forum like this one. Your writing would be more effective if it was less prolix.
One of these is unlike the other.
https://www.usnews.com/best-graduate-schools/top-science-sch...
https://tfetimes.com/best-computer-science-program-rankings/
Fun fact, the bookstore wasn't allowed to make shirts that said CMU for some time because Central Michigan University had a problem with that. Eventually they came to an arrangement and now you can buy CMU shirts at Carnegie Mellon.
Pthththth.
Go Bears!!! ;)
https://www.youtube.com/watch?v=mfebpLfAt8g
https://www.dailycal.org/2019/11/22/a-deep-dive-into-uc-berk...
https://news.ycombinator.com/item?id=24351695
"4.2 > V" BSD -vs- System V, X-Wing / Death Star Poster
https://raw.githubusercontent.com/ashaferian/Drive/master/Mt...
Funny poster, but that is not even an X-Wing. It looks more like a mashup of a BSG Viper with two WH40k Tau Burst Cannons strapped on (the older version, with the naked barrels).
One would hope that PhD students at the top-rated US universities would be able to tell the difference.