Isn’t this the guy that is famous for being admitted to a PhD program with an exceptionally low GPA?
If so, why is he the exception and why aren’t more PhD programs looking for non-traditional talent?
Edit: I read his blog post. It gave me more insight. It looks possible for people with those sort of grades to be admitted even today, but they seem to need a cheerleader on the inside that will help them.
I graduated with a Comp.Sci bachelor with a similar GPA. Spent around 10 years as a programmer in the industry and came back to get into the masters program. I was almost laughed off (a good thing) stating I would have to do another bachelor.
Sold my house, got rid of all my stuff and enrolled in pure math bachelor's. Best decision of my life.
I though that in my mid 30s with a lot more discipline, being able to work longer and harder would've made up for the shit grades. I'm extremely thankful to the department head for waving me off like that because it would've been a disaster.
Clearly Jeff could handle himself, however I think one must really know what they're doing. I thought I'd end up doing the 3-year math program in 2... it'll turn out to be 3.5 with good grades this time around. After that I know I'll be very solid for master/phd programs.
> I graduated with a Comp.Sci bachelor with a similar GPA. Spent around 10 years as a programmer in the industry and came back to get into the masters program. I was almost laughed off (a good thing) stating I would have to do another bachelor.
Can you expand on this? I'm similar right now. Want to go back and get my masters. Graduated in CS about 12 years ago. Mid 30's. What do you mean about having to do another bachelor? I have not heard this before.
> Sold my house, got rid of all my stuff and enrolled in pure math bachelor's. Best decision of my life.
I recently just bought a condo downtown. One because I never plan to move again and rent goes up at least 3-5%/year and two because I can easily rent it out if I do end up leaving. The University of Minnesota is only a mile away from me at the moment. Mortgage is probably a little high for a grad student/TA salary considering I'm making well in to the 6 digits right now. This is my only debt and all my stuff is not much. Mostly things I can't really sell for much now or are required to work and live (computer, desk, monitor, clothes, eat, sleep). I have a couch and a TV to relax outside of work.
My GPA was 2.37 and I would've had to take so many classes to increase my GPA up to the cut off point of 2.7 (B-). With 90 credits worth of classes, you'd need 90 credits at 3.0 (B average) to make up for the difference, or less if you get better grades and that means you're on the low end of what they accept.
Now I presume if you did 2-3 semester with a B+/A- GPA, they would take that as testifying you can handle the masters and let you in, however I'm enjoying math way too much right now and there are several classes (Algebra 1-3, Topology, Differential Equations and several classes in statistics I intend to pick in the option block) that will come in handy if I head into ML masters I'm also interested in Type Theory and all of Discrete Mathematics. Also, I feel math is a great program to improve at general problem-solving.
The first semester was really challenging, some classes I had in fact already done (Calculus 1 and Linear Algebra) turned out to be surprisingly difficult. I'd say around halfway through the second semester I felt I had gotten my younger brain back.
I don't have expensive habits, have two restaurants meal a week, own a car and live at the university residences (rent is 390$/month, parking 850/year). My cost of life seems to be around 12k CAD/year and am Canadian (tuition costs are 1600/semester) so I really have no idea how much you'd have to have saved up if you are in the USA.
My GPA 12 years ago when I finished was 3.5 I think. There about anyway. I figured you were referring more to the length of time you were out of school, not your GPA.
My mortgage is far north of $390, haha. I don't have or need a car though. I lease out my parking spot for $150/month or more. If I go and get accepted I'll have either work pay or try and see about scholarship or something. My brother and best friend got free rides through their master and Phd.
Jeff, I'd be curious to know. Going through TAOCP is on my lifetime-to-do and feel I am getting close to tackling it again however I have no time right now. There's also so many other things I want to go over (some higher order logic, TAPL, PFPL, the Software Foundations books, compiler design and probably won't pass up some category theory being abstract algebra seems accessible to me).
Do you feel TAOCP is worth the time investment or should I just forget about it and tackle something like your book and spend the rest of my time on other topics?
I'm not sure I can answer that question for anyone but myself. I've worked through quite a few pieces of TAOCP when I've needed to understand a particular topic, but I always find that I lose interest.
But then I've never been able to learn anything by just reading. I always have to have a target problem in front of me, and then I'll read (and get frustrated by) every book ever written to figure out the best way to think about that problem. (Which means I've read a few dozen pages from hundreds of books, and I have pretty huge gaps in my math background -- abstract algebra and category theory being two big examples.)
For some target problems, TOACP has been incredibly helpful, but for most of them it really hasn't. Knuth and I just care about different things.
For the same reason, I can't recommend that anyone work through EVERY problem in my book, either. Find the parts that are interesting and/or useful to you, and work on those. If you get tired or frustrated, work on something else; maybe you'll discover another reason to pick up my book again later. Or not.
Climbing the mountain is much more rewarding than studying the trail map.
As you've already got an answer from Jeff, I thought it might not hurt to add an additional one. IMO, you should ask yourself why you want to read TAOCP; doing it just because everyone recommends it is probably not worthwhile. Read it if you find the material or presentation interesting.
IMO one can think of each chapter (only 6 completed so far) or even each major section of TAOCP as a very deep book/monograph on that specialized topic. The writing is clear and delightful (IMO), but each of them does tend to go rather deep, more than you may care to know about that topic — so each page will require a fair bit of attention; it's not easy going. You'll get an in-depth understanding of a narrow sliver of topics.
Why not read a few of the newer sections and see if you'd like to read more in the same style? Knuth has been putting draft versions online, and they are collected here: http://www.cs.utsa.edu/~wagner/knuth/ — for example, you could read Pre-Fascicle 3B, which is on generating all [number-theoretic, or set] partitions, or Pre-Fascicle 1B, which is on a fascinating (and little-known) data structure called Binary Decision Diagrams. He uses these to solve many interesting problems, different from the focus of typical algorithms books (which would probably dismiss these methods as “brute-force”, as they don't affect the asymptotic complexity but do affect what's practical to do on real computers).
There was a small study that came out recently on biorxiv, with 29 PhD students with GREs all over the place -
>GRE scores, while collected at admission, were not used or consulted for admissions decisions and comprise the full range of percentiles from 1% to 91%. We report on the 29 students recruited to the Vanderbilt IMSD from 2007-2011 who have completed the program as of summer 2017. While the data set is not large, the predictive trends between GRE and long-term graduate outcomes (publications, first author publications, time to degree, predoctoral fellowship awards, and faculty evaluations) are remarkably null and there is sufficient precision to rule out even mild relationships between GRE and these outcomes. Career outcomes are encouraging; many students are in postdocs, and the rest are in stage-appropriate career environments for such a cohort, including tenure track faculty, biotech and entrepreneurship careers.
Now a GRE isn't a GPA (one's a specific test, one is your grades' average), but both are being used for university admissions, and I reckon both are not good predictors.
Admissions committees are looking for evidence of future success. Admitting applicants with spotty (not merely "nontraditional") backgrounds is risky -- they might be a diamond in the rough, or they might really be a weak student. And (at least departments like mine) there are far too many applicants with stellar backgrounds to justify taking that risk.
- Work on, or help out with, a research project of a professor/established researcher in your field of interest.
- Related, getting authorship (first or otherwise) for an academic publication as an undergraduate is a promising signal of future research success.
- Also related, having great references from undergrad professors who are involved in research.
- Connect with faculty in the PhD program you're interested in.
Context: Got bad grades in undergrad courses related to my particular specialty (digital audio signal processing), but also did all of these, and was admitted to a top PhD program in my field.
Also, a lot of people in my program had success with this one:
- Applying for a masters in a related field or at a department that also has a PhD program that you're interested in, and having success there.
- Own your past mistakes. They happened. Don't pretend they didn't. Figure out the underlying cause of those mistakes, and gather EVIDENCE that you've resolved that cause.
(In my case, I was a LAZY undergrad. I'd never had to work in high school, and so I didn't know how to work in college. And then I got a real job, and it was either do the damn work or it'll be there tomorrow only the boss will be there in my office wondering why the hell I'm costing the company hundreds of thousands of dollars a day and why can't you just get this shit DONE already. And so when I applied for grad school the second time I had "smart but lazy" letters from my old professors, stellar GRE scores, and "smart and works hard" letters from my managers.)
- APPLY WIDELY. You are at a significant disadvantage compared to other students with stronger backgrounds. Do not imagine that your passion and good intentions and maturity are enough to get you into the top programs, or even into any particular program. You're playing a lottery that's stacked against you; buy more tickets.
That's very good advice! One point in particular resonated with me as it seems to be applicable to science in general:
> You're playing a lottery that's stacked against you; buy more tickets.
I take this to mean that is makes more sense to apply for many things, re-submit publications often (not without considering the feedback of reviewers, obviously), and so on.
Would you take a risk on someone that graduated with below a 3.0 GPA with Cs in their math/computer science courses, but went back to school, took some core CS courses, aced them, got some research experience, and then applied to PhD programs?
There are a lot more people going to university these days, and grades are inflated. That's the short story. If you have to look at many many applications for a spot at a top university you are more likely to choose someone who is a "traditional" talent.
Are PhD degrees required to do machine learning research? Or can someone be completely self-taught and contribute to the field as much as those with PhDs?
Is vocational training and an apprenticeship necessary to be a master carpenter? No, it is not: but it makes it far easier to become one.
And that's what a PhD is: it's a vocational degree coupled with an apprenticeship. It's not like a BS or MS: it's ultimately not about courses or grades. Rather, you are attaching yourself, like a barnacle, to a researcher -- the professor -- so that you can learn how to be a researcher yourself. Eventually you step out from under her wing and do some research on your own, and that forms your thesis work.
You could learn to do research in a specialized field without going through this process -- and I know several who have -- but you'll probably discover along the way that science or engineering research is an apprenticeship field for a very good reason.
Conferences / other publication venues don't require you have a PhD to submit to them. However, figuring out what a good novel research problem is and solving it rigorously is much much easier in an academic environment --- both because you are financially supported to work on said problems and because you have a community of people to tutor you and give you feedback.
I think the financial part is important. Funding in academia is interesting to say the least. Especially when it comes to grant funding, which can often have interesting requirements that are impossible to have privately or as an individual.
note that a PhD in a computer science department is going to be much less bad than the humanities PhDs described in the article.
- they are fully funded, stipends are somewhat higher (at least not total starvation level), and there is the opportunity for lucrative industry internships
- students in general have leverage, because it's possible to leave and go get a high paying engineering job
- there's a bias towards continuing in academia but it's not taboo to go to industry; you don't have to lie about this.
- for a combination of these and other reasons, i don't think an abusive culture has set in: professors don't feel like they need to haze their students.
i don't mean to imply that everything is beautiful and perfect but it is way less culty.
I'm about to finish my masters in CS, where I focused in machine learning. While I don't research at my school, my current job is a software/ML engineer in a machine learning focused team; I consider my role between a software engineer and data scientist.
While I don't do academic research, I spend a non-trivial amount of time reading papers and keeping abreast of new research that is relevant to my job. If the research seems relevant enough, I'll dig deeper and implement the research into a workable model. The software engineering portion really kicks in when I productionize this model and make it a usable product.
So in regards to your question, if you want to do fundamental, academic research then go for your PhD. For the kind of work I do currently, then it's more about the group you work with in industry.
If so, why is he the exception and why aren’t more PhD programs looking for non-traditional talent?
Edit: I read his blog post. It gave me more insight. It looks possible for people with those sort of grades to be admitted even today, but they seem to need a cheerleader on the inside that will help them.