trynagetby
-
Posts
106 -
Joined
-
Last visited
Posts posted by trynagetby
-
-
-
applied statistics masters programs are not very competitive. If you get a good GRE score you can get into a few applied statistics professional/terminal masters. You might not be able to get into Stanford MS&Engineering, but you can definitely get into a few decent ones.
-
If you're worried about Duke being too Bayesian, I'd consider the fact that Duke's becoming significantly less Bayesian recently. Fan Li does causal inference research that's pretty non-bayesian, Alex Volfovsky works with Cynthia Rudin in the CS department on pretty straight up ML research, Jason Xu does mainly ML, Eric Laber (Susan Murhpy's student) does Susan Murphy type stuff.
So if you're not a hardcore Bayesian there's still a lot of flexibility.
-
I think your list is good with CMU will be the toughest nut to crack there. If you do well in Real Analysis, I think you'd be a serious consideration anywhere UW and below on the U.S News rankings (not to say you shouldn't apply higher, esp given diversity considerations). Also if you're seriously interested in Bayesian stuff, the obvious question is why not apply to Duke/UT Austin?
-
If you look at the number of applicants for schools that release data for previous years (Duke, Cornell for example) the number of applicants in the 2021-2022 admission cycle was very much within normal variation. So if the beginning of the pandemic didn't budge numbers too much, I don't think that this years numbers will depart from previous years very much either (aka there should not be a flood of applicants this year).
Anecdotally while there are people who delayed applications because of Covid, in my experience there are also people moved their application up a year for various reasons (I did as well as some other people I know). So it's kind of a wash.
- Nothalfgood and Cophysneurec
-
1
-
1
-
I think biology courses for biostat is borderline completely irrelevant.
With regards to going into non-bio-related fields I don't think it matters that much if you plan it right.
To be a finance quant at like a prop shot (Jane Street/Optiver/SIG) they literally only care about how well you ace the hard probability quant interviews. For other finance positions the name brand of the institution matters more. But I think doing research in related field (time series, stochastic procceses, applied probability) does help a lot, which could be slightly more difficult at a biostat department.
For tech/institutional research it depends on what you research. If you're interested in causal inference and want to do that in industry biostat programs should be great to get top industry research jobs. But if you end up resarching very niche methods for flow cytometry or phylogentics or something it'll be difficult to get anything beyond your generic data science job at a tech company (which still pays lucratively ofc). Ofc if you're researching niche methods for flow cytometry or phylogentics you'd likely be working at an actual pharma/biotech company lol
-
If the recommender writer was on the admission committee for Wharton Stat I'd listen to him. But if he was on the admission committee for Penn finance I'd stick with the prevailing wisdom that Analysis I is probably the first grade that Stat AdCom look for on the transcript. I've seen people with diversity demographics or nontraditional/funky profiles get by with an eh grade with Real Analysis (or if they're EE or Physics Undergrad) but you definitely don't fall into that category.
Good recommendations will go very far tho. I'd definitely encourage you to apply all the schools you'd want to go to if you can afford them, you never know what could happen. I definitely slightly regret not applying to enough *reach schools during my cycle.
If you're legitimately interested in Biostat, I'd apply to like all the top ones where you think you'd be happy because I think you have a chance at all of them. Harvard, JHU, UW will be hard ofc but it wouldn't be impossible. The other good schools would be like Columbia, UNC, Umich, Penn, Duke etc...
-
Those schools are probably going to be VERY difficult. Still recommend applying to them if your rec writers are okay with them and you have the money. Additionally the jump from the Chicago/CMU/Wharton/UWashington/Columbia competitiveness tier to penn biostat tier is pretty dramatic. I'd add in schools like NCSU, UCLA, Winsconsin, UNC Biostat, Mich Biostat. IMO you have a much better shot at top Biostat programs than Stat programs given your good performance in stat classes.
Also important note is if the graduate classes you mention are PhD or Masters. Masters graduate statistics classes are fairly meaningless in stat PhD admissions (as in they will not alleviate at all lack of advanced math). PhD graduate classes are a big value add (in that they can alleviate lack of advanced math).
You should check out my profile to get an upperbound of what would be realistic. I was dinged at CMU and Waitlisted at UT Austin despite having more math.
Unfortunately as an International Asian Male you're in the most competitive possible demographic category. If you don't get at least an A- in real analysis Wharton, CMU, Yale, Columbia, and Chicago are likely out of reach. If you're feeling good about your calculus a 80%+ on the Math subject test GRE would be extremely helpful.
-
If you're domestic and not strapped for cash I'd advise adding schools on the upper end of rankings if you're interested in their programs (think Uwashington, Duke, etc...). This advice especially applies if you're very close with those two professors and expect good LOR.
Edit: Not sure when all the application deadlines are so apologies if they have already passed.
-
There's nothing wrong with an LAC! I think your list is still reasonable , if pretty top heavy. The PhD courses definitely help a lot and I think that's a reason why you have a good shot at those programs, but generally the rule is the more math you have as an undergrad the better.
Honestly, the biggest factor will be the strength of your recommendation letters. If you think your letters are top-notch your list is probably good. Take my advice with a grain of salt as I only applied to 2 biostat schools, and got into 1.
-
List seems pretty reasonable. Just posted to say that if you're interested in doing research/pursuing PhD afterwards DO NOT even apply Columbia Stat MS. Also your GRE is probably fine.
-
If you've had a lot of advanced mathematics background (Real Analysis I, II, Proof-Based Lin-Alg, + one more advanced class like Analysis of Algs, Abstract Algebra, Numerical analysis etc.. ) those schools are definitely all well within reach. If you've just had Real Analysis or even no Real Analysis you have a dece chance at all those schools but a clean reject from all of them is very well in the realm of possibility (I don't know anything about BU).
Given your locational preference those schools pretty much cover the northeast. I'd throw in JHU too if Maryland isn't too south for you. Safe(r) schools would be like UNC, Umich, Duke. But those are pretty out your geographical preference.
-
2 hours ago, RandomMathGuy said:
Thanks a lot! What about Stony Brook and Pittsburg? Are they within reach?
I think you have a pretty good chance to get into them. I'm don't know much about those programs, but in general for domestic students the competitiveness of PhD programs falls drastically once you get out of the union of top 30 Stat Schools and top 30 general schools.
-
If you got A's in Real Analysis/Applied Analysis (which I'm guessing is like baby functional analysis?) I think you should apply to the range of NCSU/Wisconsin Madison/UIUC/Rice/Texas A&M and below. NCSU/Wisconsin might be a little of a reach but its pretty possible :). I'd also think about applying to programs like Umichigans Biostat Masters (which is funded) to give yourself a leg-up before going PhD. If Biostats is an option, UNC biostats and below should be a good chance. Good Luck!
-
To be honest, I think applying to Finance Departments/Business schools/Operations Research will be better of a fit.
-
I think you're tremendously underselling yourself. That's a whole lot of math, and not having life sciences coursework doesn't matter at all (I got an offer from Harvard Biostat and I had zero sciences background). Biostat at UW-Harvard-JHU might be too tough to crack, but I think an application to like UNC-Michigan would be worth it and I think you are very competitive at a place like Minnesota. For Statistics onwards I'd apply to a few schools at the level of Duke-UW-Michigan then work your way down to some safeties with the bulk of your applications in the USNR 20-40 range (I think Rice is a really safety school).
Good luck!
-
6 minutes ago, aeiou12345 said:
I apologize, I am somewhat new to this forum -- how can I see your application profile? I can, like, see your previous posts but can't seem to see your stats anywhere. Thank you very much for your help, by the way!
No worries, you can scroll down to my post in this thread:
Good Luck!
-
Check out my profile from way back which I think is similar to yours but less OR focused. I also applied having no idea what OR was like and I got into Northwestern and GaTech ML (ISYE) with special fellowships for both. I only applied to the two because my main interests was Statistics PhDs. I think you have a good shot at all of them except MIT because, well, MIT (still worth applying IMO tho)
-
Harvard, Berkley, UW Stats all have at least 3 very top/rising star type people doing causal inference research. An important distinction you have to make is whether you want to do "classical" causal inference (propensity scores, average treatment effects, instrumental variables, potential outcomes framework) or "modern" causal inference (dags,judea pearl causal discovery, reinforcement learning, adaptive designs etc...). Both are pretty hot right now but the flavor of research is extremely different.
-
3 hours ago, DanielWarlock said:
I don't think it's impossible but rather depend on OP's "math maturity" and how these classes are taught at that particular year. OP is an actuarial not a musician. Actuarial studies is a mathematics degree at Waterloo. They do already know a great deal about theoretical things like SDEs.
With certain math maturity, one can certainly take these de factor self-contained classes that tend to be taught from scratch even at graduate level. Every year, a couple sophomores or even freshmen take grad probability and grad real analysis with us. Last year we even had a junior as teach fellow for graduate real analysis class. So it is very likely OP would be able to take these classes. If not, OP will always have a chance to drop out. I don't think it's a big deal.
If the OP has taken the equivalent of Math 55 he/she should definitely go for it. Harvard's the odd case as their undergraduate pure math first year math class literally has its own wikipedia page lol
-
I'd say depends on the professor/class. If your Real Analysis class was super basic and your Complex Analysis class was like Stein and Shakarchi or Papa Rudin, I'd pick Complex Analysis. But if they're similar and you got into a competitive REU, you know apriori that the rec has to be somewhat decent right?
-
2 hours ago, untzkatz said:
Sounds like what you are getting at is that the coursework in statistics/biostatistics departments is heavily foundational classical stats, but the research does more modern things and combines it with the inferential aspects? Whereas NYU DS for example seems to go right into the statistical ML/DL and bayesian network type stuff.
I would be interested in Comp Neuro too since you brought that up. Did you do that stuff in undergrad-it seems very advanced for undergrad level. I agree the good thing about an Algorithms course is that even if I don't do a PhD, it can still help to get through interviews at tech companies since that stuff is tested in Leetcode and so on. And still improves general programming skills beyond just numerical computation. That is why I have been leaning towards doing it. As far as the frameworks though, I'm pretty sure most PhD students doing DL are using PyTorch and so aren't implementing various data structures or autograd from scratch. I've seen arxiv github code and it still often follows the formulaic subclassing nn.Module to make a layer, then having __init__ and forward() and so on. And making a Dataset class and Dataloader.
I'd recommend watching this NIPs lecture from Robert Tibirishani
to get an idea of the type of problem that Statisticians are interested in and the inference/prediction ML distinction. It just seems you're not interested in statistical inference. There's nothing outdated about inference. It's just a goal that the statistics field is interested in, and the tools to achieve that tool are always evolving.
I was a fairly mathematically/computationally mature undergraduate working with a super nice prof who works with undegrads.
That type of program does seem well suited to your interests. But UCSD ECE is an extremely hard department to get into (as well as as NYU DS). You should still apply but hedge your bets with applications to bioinformatics/OR/BME programs as well.
-
On 5/4/2021 at 2:47 PM, untzkatz said:
I don't know if I am interested in pipelines like the software/ML engineering sense of the term, but I like both methods+applications of ML/DL. Sometimes, people develop methods for specific applications. My undergrad btw actually was in BE which you listed, although it is far too broad and I wish I did something like applied math ugrad.
My Biostat program covered those inferential things at MS level in the 1st year, and 2nd year we had stuff on Survival+GLM+GLMM (these were combined with PhD students, I got an A in survival/GLMM but a B+ in GLM) as well as the electives which were the ML/signals/time series classes I mentioned (those were all As). Yea, the higher ranked programs I would expect to be more modern. I'm not too interested in doing the Fisher/Neymann Pearson inference stuff all over again though, but NYU DS inference & representation course on the graph models does look interesting as its a more modern spin on it.
Didn't realize DS&Algs would be needed though, I'm not sure where things like heaps, linked lists, dynamic programming etc come up in ML at all, but I also took ML in a stat department not a CS one. Technically I know NNs for example make use of dynamic programming when using autograd to cache the gradients, but thats in internal detail you don't need to worry about when using high level programming languages/frameworks like Julia's Flux or the TF/PyTorch frameworks. The only computational complexity stuff we did was related to matrix decomps in computational stat. Between Real Analysis and an Algorithms class what would be better? I don't think I can take both.
Idk man, if you think you can get what you want out of a Statistics PhD then go for it. But it really doesn't sound like you'd enjoy the curriculum or research focus. Im quoting the first line out of the syllabus for the last course in Stanfords Statistical Inference Sequence:
- Testing problems in high dimensions: sparse alternatives (needle in a haystack) and nonsparse alternatives, Bonferroni's method, Fisher's test, ANOVA, higher criticism.
Even CMU which is really MLey out of all the statistics departments requires to review topics like
- simple linear regression, ordinary least squares and weighted least squares, the geometry of least squares, quadratic forms, F tests and ANOVA tables, interval estimation, minimax theory, hypothesis testing, data reduction, convergence concepts
You are interested in ML, but maybe not from a statistical perspective. Statisticians do all the things you're talking about but you absolutely have to prove inferential properties and understanding the basic foundations of hypothesis testing is necessary.
Honestly, you should check out programs like https://bioinformatics.gatech.edu/ through the ISYE (read:OR).
I think the problem here is "using high level programming languages/frameworks like Julia's Flux or the TF/PyTorch frameworks." When you're doing academic research you can't be constrained to pre-packaged stuff that everyone has access to. You have to do something novel and new to data which no one has before. That will inevitably involve implementing something from scratch. For example in my computational neuroscience research for a statistics prof at a top school, I once had to find the cluster of vectors lying on a sphere that maximized the sum of projection onto them by a given vector with certain contraints. How do you go about this the fastest, what data structure do you use, can you approximate? DS professors will want to know you have the tools to think about this.
It's difficult to say what you should take. Analysis of Algorithms will be useful for jobs and might be enough to get your foot in the door some CS/OR places. Analysis I will be the only way you get into decent Statistics/Bio-statistics programs.
Honestly I think you need to read dissertations from places like UWashington/Harvard/JHU biostat and really make sure you're not interested. You seem to be really hung up on hypothesis testing and asymptotics being boring when the concepts are kinda the core of Statistics.
-
13 hours ago, untzkatz said:
I see, yea I am not interested in overall CS though. I feel like I only like this narrow ML/DL area and to me it seemed like stats. So seeing that NYU DS you can more or less just focus on that area looks appealing. They do a lot of MRI research in the biomedical track too, which seems to be more applied statistics based than CS. I do agree more math background would help but I was in a different biomedical field in my undergrad, so can’t do much now. I could potentially sign up for one of either Real Analysis or Data Structures&Algs for the summer though, which I am considering.
Both would be important especially for NYU datascience, especially a formal AoA class (Dynamic programming, Graph Algorithms, basic NP-Completeness proofs) because professors need to know that you can reason rigorously about the complexity and correctness of novel algorithms.
13 hours ago, untzkatz said:Honestly I never really thought I would like ML/DL back in undergrad because I didn’t really know what it was and thought it was some insane CS thing but 2 stat ML (on supervised+unsupervised learning) classes in my MS I was wowed and I got convinced that ML is stats. And computational statistics (which had a bit of numerical analysis, but mostly matrix decomps/GD/MCMC/EM) had some as well. I also had a signal processing (special topics) stats course I really liked on FFTs, which was actually invented by Tukey, and I liked that too. Time series as well but my TS elective course was undergrad level. There was much less asymptotic stuff in these areas and it was more like “show gradient descent on convex functions converges” which is more optimization. So it seems weird to me that all this stuff isn’t considered statistics, perhaps I went to a more modern department after all. Deep Learning also we had a little bit on it from the GLM/GAM perspective. And I always liked GLMs and regression, so seeing that ML and DL boiled down to that got me interested in it. And regularization too, like how people nowadays are incorporating different kinds of penalties for domain specific problems is interesting. The whole double descent thing in DL to me seems like statistics, at least the way Dr Witten explained it with GAMs and regularization. VAEs for example seem to be heavily statistical would fall under probabilistic modeling of multivariate distributions, letting you generate new data based on the latent space. I don’t think asymptotics, inference, and p values necessarily define the field.
It seems like you're more interested in developing bioinformatic pipelines that use ML techniques than developing specific ML methods (at least in statistical sense). In modern statistics, optimization and modern techniques are important, but ultimately you have to prove that these techniques give good inference. Focus on modeling, inference and estimation are what differentiates Stats from other fields. While there are people in Stats who do the work with NN/VAEs/GAMs that you're talking about, they're the exception to the rule. People like Liam Paninski and John Cunningham from Columbia who do that type of stuff in statistics departments do so mainly in service to a field like neuroscience and I'm not really sure why their primary appointment is in the statistics department. I think you should seriously think about a PhD in like EE/Bioinformatics/Bioengineering/Computational Neuroscience if you're really not interested in inference and estimation.
Idk what department you went to, but Stats PhD departments from Stanford to Berkley to Uchicago to Duke to Washington all require students to take heavy course load in classical inference (linear models, hypothesis testing, ANOVA) and asymptotic statistics. Albeit, the top departments like Stanford, CMU, Chicago, and Berkely put a very high dimensional spin on this, but its fundamentally the same goal. Correct me if I'm wrong, but I think all those departments are fairly modern. If every department is teaching these courses, I would think that they're fundamental to statistics. Lower ranked programs are even more focused on "classical statistics". If you attend these programs without interest in inference, you'll at least be bored for a year.
Also take my rambling with a grain of salt, I'm an incoming grad student and I don't even hold a masters.
Rankings before the Biostat/Stat separation
in Mathematics and Statistics
Posted
I think the point of the separation is that Biostatistics and Statistics programs are pretty different worlds and the comparison between departments in the two fields was not particularly productive.