UMich EE:Systems vs CMU ECE

[DSPCSbuff]

Hi All,

I got an admit from cmu ece last week. Earlier this week I got an unofficial informal sort of admit from UMich EE:Systems. I am in a real dilemma as to select which one of these two. My area of interest being Signal Processing..Statistical Signal Processing and Compressed Sensing to be precise. With no communication from UCB(I consider it to be a certain reject!) and a probable reject from Princeton, these two boil down to be my top choices. 

 

With CMU though, I have been working with my POI for over an year now. I did a REU with him last summer. At the same instance, he is young and new and an Assistant Professor. I really liked working with him. The same goes for my UMich POI, but he's may be a year or two more experienced in comparison to my CMU POI.

 

I don't have much idea as to where do CMU and UMich figure in ece rankings. Should I take that into consideration too?

 

I am having a real hard time choosing between the two and I will appreciate any help I get for the same.

 

Thanks a lot!

Edited February 13, 2013 by Anit Sahu

[charlesjeon]

If you're into compressed sensing, isn't Rice the best place to go?

 

They have their own compressed sensing group led by Baraniuk.

 

Why don't you try waiting for Rice/GaTech? GaTech has Romberg too.

[DSPCSbuff]

Well Rice has a very good compressed sensing group for sure led by Baranuik, but there work these days is pretty much aimed at applying compressed sensing to images and its processing. GaTech has Romberg/Davenport and given they are Rice DSP alumni they work on similar lines. With CMU and UMich I get to work on more theoretical signal processing and compressed sensing rather than applying it to things. At least this is what I think. Thanks anyways!

[charlesjeon]

That's true. I'm not sure if there are more theoretical CS research that is going on these days. I know it's still a very HOT topic, but not sure if they will still be the next big thing.

 

It's cool to see this thread because I'm also interested in the signal processing area (image/video processing). I'm still waiting for Michigan, but got offers from Texas/GaTech.

 

Have you tried contacting Prof. Hero?  I've contacted him back in October but he told me that he was unsure if he was going to take any students.

Edited February 14, 2013 by charlesjeon

[DSPCSbuff]

I am more into signal processing theory rather than video/image processing. My area of interest would rather be Applied Maths and Signal Processing. I did contact Prof. Hero but never heard back from him. I contacted my POI in UMich who works on Random Matrix Theory and its applications. With CS more and more pure mathematicians are coming into it, many of them even cite it as their research interest. So, the field is progressing at an unbelievable pace.

 

Nice to find somebody having similar interests.

[Deadmeat]

What are your thoughts on the non academic side of things? CMU and Michigan are completely different types of Universities when it comes to the student population size, athletics, school events, etc. I just said this in another thread, but personally I love Big Ten programs, and you should give thought to what you want out of the University as whole outside of the program. You might find your answer that way.

[Nick PhD EE]

Hi Anit,

 

I would suggest three points:

1) If you know your prof at CMU and can work with him then I would recommend u to go to CMU. My gut feeling is PhD is a long journey and if you have healthy relationships with your prof then the journey becomes smooth and memorable. On the contrary if you find it hard to work with your CMU prof then this is the time to make a switch. 

 

2) Also, if you are like me, who badly wants to work with a particular research group (I always wanted to be a part of the wireless research lab at UCB) then UMICH might be a good choice. You can join there and take a transfer after a year (or semester). I feel it is better to ditch a new relationship than an established one.

 

3) You have not mentioned funding. It is one of the most imp metric to be considered. I would definitely go for a Univ which offers fellowship versus an RA/TA position. As pursuing PhD is a new experience, you would want to dedicate all the time for it. At least for the first year.

 

These were my personal opinion. I just hope it might be of some help. 

Cheers,

Nick

[tarrman]

You and I are going for the exact same field: I'm also interested solely in the theoretical/mathematical side of signal processing, so maybe I can offer some advice.

For me, I'm more interested in the math than anything. I'm willing to work in an ECE research field so long as there's a lot of math and theory. If you're the same way, there are other fields to consider within ECE that offer the same level of advanced math as compressed sensing, such as machine learning. It's a rapidly growing field that's gaining a lot of attention, and a lot of the theoretical work is still being developed. Many professors who work in compressed sensing/sparse approximation also do work in machine learning. If your mind is set on compressed sensing, however, ignore everything I just said. Otherwise it might be helpful to see which schools have a good bit of professors working in these fields to see which one has a stronger program. I know CMU is well reputed for its work in machine learning.

 

Also, be wary of taking on an adviser who is still only an Assistant Prof. Their jobs are much less stable and are more likely to transfer to another school (statistically). He'll also be up for tenure during the time you work with him, which may take him away from his mentoring duties. This isn't a guarantee though, and many people have done their theses with an assistant professor without any issues. But if you're looking to get into academia after getting your Ph.D., it'll be much more difficult doing so having worked with someone who isn't very well known (but not impossible).

I really don't think you could go wrong here with either choice. Basically it should come down to which school is doing research you're more interested in, which gives a better financial aid package, and which is in a better location (for your liking). Hope some of this helped!

P.S. I'm visiting Rice this week, and I'm planning on meeting with Dr. Baraniuk to talk about his current research plans. If you're interested, I'd be happy to relay an information I get!

[DSPCSbuff]

@Nick Both my admits come with full financial funding. CMU ECE on one hand doesn't hand out fellowships, but gives RAships worth $29k per year which I think is pretty good considering Pittsburgh. Same goes for UMich too. I am very comfortable working with my CMU POI. I like the UMich POI's work a touch better may be speaking maths wise. So, in a big big dilemma. Thanks a lot for your inputs anyways

 

@Tarman.. It feels really good to find people around here interested in Applied Math and Signal Processing. Both my POI's are assistant professors, so nothing much to choose that way. I will look to go for a Post-doc position with a big shot, if at all I intend to join academia once I get done with my PhD. Yeah, machine learning in signal processing is coming up big time. IEEE has already come up with a society for that. But about the Assistant Professor vs Big Shot debate, I got a bird's eye view of it during my REU at CMU. During my REU my POI used to spend a hell lot of time having academic interactions with me, but saw a big shot meeting his grad students with a frequency of once in a month. Though with all the added benefits of working with a big shot for post-PhD life, I think I would like the interactions more as long as they don't dictate my work completely.

 

About the tenure, may be I am not potent enough to comment, but the rate at which my CMU POI is going, I don't see him not getting a tenure.

 

It would be really nice to know about Prof. Baranuik's research plans. I am also looking forward to the Rice decision and have obviously applied to him. Its just that I don't like the part because they look to apply compressed sensing at places. Though theory does come up with it, I like core signal processing better. Thanks a lot again

[charlesjeon]

Hey guys,

 

I don't want to hijack this thread, but what do you guys think of Cornell in signal processing/inf theory?

 

I'm deciding between UT, GaTech and Cornell, but need some opinions.

 

Personally, I would prefer Cornell, but then again, UT and GaTech are both excellent schools in signal processing.

[tarrman]

Hey guys,

 

I don't want to hijack this thread, but what do you guys think of Cornell in signal processing/inf theory?

 

I'm deciding between UT, GaTech and Cornell, but need some opinions.

 

Personally, I would prefer Cornell, but then again, UT and GaTech are both excellent schools in signal processing.

 

It all really depends on what you want to do. I'm currently an undergrad at Georgia Tech, and I can tell you the the DSP department here is very broad and strong. At least five professors did their Ph.D. at Rice (also has a great DSP problem). Rozell, Romberg, and Davenport all work on the theoretical side of signal processing. If you're interested in speech, there are a good number of professors as well (Clements, Lee). There are a few other professors working in networks and also in acoustic/audio processing. There are even some professors working in radar (and GTRI is a really great resource if you're interested in radar SP). A word of caution, however, if you're interested in academia, Georgia Tech is primarily aimed at industry prep at both the undergraduate and graduate level, and very few graduates go on to become professors. But there are always graduates who do, and some of the professors are well known enough to help establish yourself. The dept. is also very large, so if you're looking to get to know everyone and develop a relationship, don't count on it at Tech. But overall I think it's a great program (I'm just looking for a change of scenery).

 

As for UT, I've been accepted there and will be visiting next Thurs. I'm not sure if you've talked to anybody there, but from what I've gathered, most students are co-advised, which allows for more felxibility in the topics you study. They also gave me a LOT of funding. And I think Austin is arguable the best city to live in in America (numerous people have told me this, in and out of Austin). As for the DSP dept. Dr. Bovik is extremely well known in his field. I'm not very familiar with the other professors, and it seems like there are a lot of young faculty members. And UT has some strong industry connections (although I think GT's are stronger). Sorry I can't give a better assessment, but I'll know more about the program next week.

 

As for Cornell, I don't really know much about SP and can't judge the strength of the DSP dept. I think I heard they have a good imaging dept. I know that Parks (well-known pioneer in DSP) worked there before retirement, so I imagine the program is decent. They strike me as the schoold doing the most theoretical research of the three. If you have the opportunity to visit, I strongly recommend taking advantage of it.

 

Hope this helps!

[tarrman]

@Nick Both my admits come with full financial funding. CMU ECE on one hand doesn't hand out fellowships, but gives RAships worth $29k per year which I think is pretty good considering Pittsburgh. Same goes for UMich too. I am very comfortable working with my CMU POI. I like the UMich POI's work a touch better may be speaking maths wise. So, in a big big dilemma. Thanks a lot for your inputs anyways

 

@Tarman.. It feels really good to find people around here interested in Applied Math and Signal Processing. Both my POI's are assistant professors, so nothing much to choose that way. I will look to go for a Post-doc position with a big shot, if at all I intend to join academia once I get done with my PhD. Yeah, machine learning in signal processing is coming up big time. IEEE has already come up with a society for that. But about the Assistant Professor vs Big Shot debate, I got a bird's eye view of it during my REU at CMU. During my REU my POI used to spend a hell lot of time having academic interactions with me, but saw a big shot meeting his grad students with a frequency of once in a month. Though with all the added benefits of working with a big shot for post-PhD life, I think I would like the interactions more as long as they don't dictate my work completely.

 

About the tenure, may be I am not potent enough to comment, but the rate at which my CMU POI is going, I don't see him not getting a tenure.

 

It would be really nice to know about Prof. Baranuik's research plans. I am also looking forward to the Rice decision and have obviously applied to him. Its just that I don't like the part because they look to apply compressed sensing at places. Though theory does come up with it, I like core signal processing better. Thanks a lot again

 

In regards to the tenure comment, most professors eventually obtain tenure. I only meant that prior to getting it, the professor may be more busy/stressed out than he normally is, which sometime can affect his ability to work with his graduate students. But you already have experience working with him and should have a good idea how he handles his own work, so this may not even be a problem for you.

 

I also had a chance to meet with Dr. Baraniuk (although very briefly). He was extremely nice and probably one of the most laid back professors you will ever meet. He was very informal, modest, and an excellent speaker. I think as a mentor he would be a lot of fun to work with. All the grad students and current ugrads said he was one of the most unique and friendliest professors on campus.

 

As for his research, he described it as purely mathematical foundation (no design whatsoever). He collaborates with many professors in the dept. of statistics and even mentioned that he was interested in having a mathmetician in his group. He did mention that he was doing a lot of work in applications, as you had mentioned, however it was all on developing the mathematical foundation for doing so. Some examples he gave were hyper-spectral imaging, A/D conversion, infrared cameras, and MRI/CT scans.

 

He's doing less applied work in machine learning and his examples of the kind of math he was doing was manifold modeling, factor analysis, and collaborative filtering. Again, as I mentioned in my previous post to you, I'm guessing the reason for this is it's a developing field. Overally I was very impressed with him and found his interests to align almost perfectly with my own.

 

My only concern is that it seemed like most of the other students at the recruitment weekend were primarily interested in his work (a lot of DSP people). However, I and one other student were the only onese primarily interested in theory. Also, the grad students mentioned the acceptance rate is only 10% Decisions should be coming out in mid-March or so (first year fully funded by fellowships for all admits).

[charlesjeon]

It all really depends on what you want to do. I'm currently an undergrad at Georgia Tech, and I can tell you the the DSP department here is very broad and strong. At least five professors did their Ph.D. at Rice (also has a great DSP problem). Rozell, Romberg, and Davenport all work on the theoretical side of signal processing. If you're interested in speech, there are a good number of professors as well (Clements, Lee). There are a few other professors working in networks and also in acoustic/audio processing. There are even some professors working in radar (and GTRI is a really great resource if you're interested in radar SP). A word of caution, however, if you're interested in academia, Georgia Tech is primarily aimed at industry prep at both the undergraduate and graduate level, and very few graduates go on to become professors. But there are always graduates who do, and some of the professors are well known enough to help establish yourself. The dept. is also very large, so if you're looking to get to know everyone and develop a relationship, don't count on it at Tech. But overall I think it's a great program (I'm just looking for a change of scenery).

 

As for UT, I've been accepted there and will be visiting next Thurs. I'm not sure if you've talked to anybody there, but from what I've gathered, most students are co-advised, which allows for more felxibility in the topics you study. They also gave me a LOT of funding. And I think Austin is arguable the best city to live in in America (numerous people have told me this, in and out of Austin). As for the DSP dept. Dr. Bovik is extremely well known in his field. I'm not very familiar with the other professors, and it seems like there are a lot of young faculty members. And UT has some strong industry connections (although I think GT's are stronger). Sorry I can't give a better assessment, but I'll know more about the program next week.

 

As for Cornell, I don't really know much about SP and can't judge the strength of the DSP dept. I think I heard they have a good imaging dept. I know that Parks (well-known pioneer in DSP) worked there before retirement, so I imagine the program is decent. They strike me as the schoold doing the most theoretical research of the three. If you have the opportunity to visit, I strongly recommend taking advantage of it.

 

Hope this helps!

 

Hey tarrman, thank you so much for your feedback. I've been in contact with Romberg/Davenport and will most likely join their group if I go to GaTech. As for UT, I've been in touch with one of the Professors at the WNCG group and just been informed that I would get a fellowship+RA. I haven't received any info yet, do you know how much the stipend was?

 

If I go to Cornell, I think I would be more on the theoretical side. I know that UT/GaTech have more industry feel to it, but then again, I guess it depends on the advisors. Anyways, thanks for the help! I wanted to visit Austin for the visit day, but I can't go

[DSPCSbuff]

In regards to the tenure comment, most professors eventually obtain tenure. I only meant that prior to getting it, the professor may be more busy/stressed out than he normally is, which sometime can affect his ability to work with his graduate students. But you already have experience working with him and should have a good idea how he handles his own work, so this may not even be a problem for you.

 

I also had a chance to meet with Dr. Baraniuk (although very briefly). He was extremely nice and probably one of the most laid back professors you will ever meet. He was very informal, modest, and an excellent speaker. I think as a mentor he would be a lot of fun to work with. All the grad students and current ugrads said he was one of the most unique and friendliest professors on campus.

 

As for his research, he described it as purely mathematical foundation (no design whatsoever). He collaborates with many professors in the dept. of statistics and even mentioned that he was interested in having a mathmetician in his group. He did mention that he was doing a lot of work in applications, as you had mentioned, however it was all on developing the mathematical foundation for doing so. Some examples he gave were hyper-spectral imaging, A/D conversion, infrared cameras, and MRI/CT scans.

 

He's doing less applied work in machine learning and his examples of the kind of math he was doing was manifold modeling, factor analysis, and collaborative filtering. Again, as I mentioned in my previous post to you, I'm guessing the reason for this is it's a developing field. Overally I was very impressed with him and found his interests to align almost perfectly with my own.

 

My only concern is that it seemed like most of the other students at the recruitment weekend were primarily interested in his work (a lot of DSP people). However, I and one other student were the only onese primarily interested in theory. Also, the grad students mentioned the acceptance rate is only 10% Decisions should be coming out in mid-March or so (first year fully funded by fellowships for all admits).

Thanks a lot for letting me know about Prof. Baranuik's research plans. Meanwhile there have been some developments on my side from UMich and CMU. In my UMich offer my POI clearly mentions of the possibility of collaborating with Prof. Hero whom you might know to be as one of the big shots in Statistical Signal Processing. During my interview with my POI as well, he mentioned about the low acceptance rate for the EE:Systems program. But their work is more or less focussed toward statistical inference, random matrix theory and its applications and information theoretic signal processing.

 

On the other hand from the CMU side, Prof. Moura could probably be my co-advisor. CMU's SP is pretty strong and to some extent is  centred around my POI and Prof. Moura. My POI in CMU clearly mentioned about the project which defines the paradigm for my PhD. It would be on adaptive sensing or rather compressed distilled sensing. As you would know, compressive sensing is non-adaptive in nature and is generally utilised for offline processing. Distilled Sensing came up very recently from the likes of Prof. Nowak et.al which provides for adaptive sensing and that too with online processing . So, the research area is really pulling me towards CMU. Though UMich's Maths department and the thought of being able to do courses in Harmonic Analysis excites me a lot. The dilemma is almost cured for now with me inching towards CMU.

 

I too got to know the same about Rice's acceptance rates from one of my friends working towards a PhD there. So, not counting on it too much at the moment.

 

P.S Would appreciate your thoughts on this.

[tarrman]

It sounds to me that the research opportunities at both schools are about equal. I'm not familiar with CMU (and consequently Dr. Moura), but I know the school carries a very prestigious reputation and that they are very strong in theory. I don't think you could go wrong in picking one school over the other. I imagine the acceptance rates at CMU (and the other schools you applied) are also comparable.

 

I think it comes down to which research sounds more interesting to you and which schools have a broader range of professors doing research that you know you'd like to be a part of. You'll be doing the same amount of math in both programs, and I'm sure you'll have the opportunity to take courses in other interesting topics, regardless of where you go (so don't worry about missing out on harmonic analysis). If you really like the advisor you worked with at CMU and the research he was working on, then I'd take the offer. It's nice to have an established relationship with an advisor you know you can work with. And CMU's program is strong enough that if you for some reason wanted a different advisor, you wouldn't have a problem finding another professor to do research with.

 

Hope this helps! I know it's very tough to make such an important decision, which is why I'm going to wait until the end of March/early April to make mine.