Author Topic: Advice on PHD topic  (Read 1740 times)

0 Members and 1 Guest are viewing this topic.

Offline trys11

  • Contributor
  • Posts: 11
  • Country: gr
Advice on PHD topic
« on: January 12, 2019, 06:18:52 am »
Greetings everyone,
I am not sure if I am posting this in the correct section of the forum but hopefully I am.
I just finished my masters degree in electrical and computer engineering and I am convinced that pursuing a phd is the way to go.

What troubles me is the subject of it.
What i am interested in is hardware design. What I have already done in various projects is program an FPGA using VHDL.
So I would like to pair that interest with another field as the subject of my phd.
I have some background in cryptography, computer vision, computer networks and some other fields as part of my basic training.
So the big question is what is the most promising subfield of these or what else would you recommend.
I am open to hearing any kind of ideas.
Also, any general tips in starting a phd would be MUCH appreciated.

Thanks in advance for your inputs
 

Offline Ice-Tea

  • Super Contributor
  • ***
  • Posts: 1293
  • Country: be
    • Freelance Hardware Engineer
Re: Advice on PHD topic
« Reply #1 on: January 12, 2019, 07:25:50 am »
Not having a PhD, I keep a small list of subjects that would fit. Having studied and having done some projects, you must have had some moments when you went "what about..., what would happen if..., I wonder..."?
eBay shop with all the gear you need!
FS: Agilent 54815A, 54825A, R&S CMU200, CRTU, SFU, SMIQ06L, SFU, HP8714B, Marconi 6201B, Lecroy WP 960,950, 9354TM, THS720P, Anritsu MG3671A 2.75G I/Q RF gen, MS8604A 8.5GHz SA
 
The following users thanked this post: trys11

Offline trys11

  • Contributor
  • Posts: 11
  • Country: gr
Re: Advice on PHD topic
« Reply #2 on: January 12, 2019, 07:51:41 am »
Not having a PhD, I keep a small list of subjects that would fit. Having studied and having done some projects, you must have had some moments when you went "what about..., what would happen if..., I wonder..."?

I am not sure if I understand your question.
 

Offline Ice-Tea

  • Super Contributor
  • ***
  • Posts: 1293
  • Country: be
    • Freelance Hardware Engineer
Re: Advice on PHD topic
« Reply #3 on: January 12, 2019, 08:57:48 am »
You have been studying and learning for years. Surely you must have come across a few questions worth answering?
eBay shop with all the gear you need!
FS: Agilent 54815A, 54825A, R&S CMU200, CRTU, SFU, SMIQ06L, SFU, HP8714B, Marconi 6201B, Lecroy WP 960,950, 9354TM, THS720P, Anritsu MG3671A 2.75G I/Q RF gen, MS8604A 8.5GHz SA
 
The following users thanked this post: trys11

Offline trys11

  • Contributor
  • Posts: 11
  • Country: gr
Re: Advice on PHD topic
« Reply #4 on: January 12, 2019, 09:16:14 am »
I guess so, but nothing comes up in my mind right now.
 

Online blueskull

  • Supporter
  • ****
  • Posts: 11077
  • Country: cn
  • Power Electronics Guy
Re: Advice on PHD topic
« Reply #5 on: January 12, 2019, 09:45:54 am »
When I did my PhD, my PI asked me to be one of the world's top 6 experts in my exact field of research for graduate.
I think you can get the point.

You don't need to go super wide unless you want to do them as a side hobby.
Your main task is to focus on one thing, and dig deep enough to be the definitive expert on that matter.

Having a broad understanding is equally important. Don't be the EE PhD that doesn't know which end of a soldering iron to grab.
Just don't let the side tracks diverting too much of your time form your main research. It's an intricate balance.

Reference: I graduated my own MS+PhD from 2014 to 2018.
 
The following users thanked this post: trys11

Offline trys11

  • Contributor
  • Posts: 11
  • Country: gr
Re: Advice on PHD topic
« Reply #6 on: January 12, 2019, 10:04:02 am »
When I did my PhD, my PI asked me to be one of the world's top 6 experts in my exact field of research for graduate.
I think you can get the point.

You don't need to go super wide unless you want to do them as a side hobby.
Your main task is to focus on one thing, and dig deep enough to be the definitive expert on that matter.

Having a broad understanding is equally important. Don't be the EE PhD that doesn't know which end of a soldering iron to grab.
Just don't let the side tracks diverting too much of your time form your main research. It's an intricate balance.

Reference: I graduated my own MS+PhD from 2014 to 2018.
Thanks a lot for your input!
Your accomplishments are remarkable 4 years for a ms+phd looks like an outstanding feat.
Most PHD programs i find last from 3 to 5 years. Also in some cases, i think you have to stay at least 4 years but i think that depends heavily on the given University.
To explain myself a bit better, I dont like doing hardware for hardware's sake, I want to do it in order to make an implementation for something (a cryptographic cipher for example).
So what I need, is to find a good theoretical part to make implementations for and I am asking help and advice in that regard.
Would it be better for me to do hardware implementations in the area of hardware security or maybe i should opt for computer vision ? Or maybe I should get into smart edge computing and make hardware implementations there ?
Thats the kind of questions I have
 

Online blueskull

  • Supporter
  • ****
  • Posts: 11077
  • Country: cn
  • Power Electronics Guy
Re: Advice on PHD topic
« Reply #7 on: January 12, 2019, 10:26:10 am »
Would it be better for me to do hardware implementations in the area of hardware security or maybe i should opt for computer vision?
Or maybe I should get into smart edge computing and make hardware implementations there?

First of all, I know a bit of all the topics, and I have friends doing all topics above. But personally I never know more enough to do research on either. I only learn enough to get jobs done for a particular need, aka, I only know the applications and how to use them, not enough to research them.

----------------------------------------------------------------------------------------------------

Crypto is fine, but you will be competing with companies like AntMiner and other ones who have lots of cash on building 28nm or even more advanced chips for the same purpose. If you do advanced crypto algorithms, you will need extra math skills, and your future jobs will be very limited.

CV has a pretty good future, and also a good immediate market AFAIK. Especially when deep learning comes into the play, CV is becoming hot again in the recent years. Lots of math models and ideas were proposed, but never implemented nor perfected to usable. Those make easy topics and publications to get you graduate quicker and land jobs quicker.

IoT is really more of a concept than a concrete theory. I would not do research on IoT if I want a solid background that can not only launch my career but also support my entire life.
 
The following users thanked this post: trys11

Offline trys11

  • Contributor
  • Posts: 11
  • Country: gr
Re: Advice on PHD topic
« Reply #8 on: January 12, 2019, 10:38:17 am »
Thanks a lot!
So from these fields you suggest the computer vision one. What do you think about networking (schedulers for example)?
Also if you have any other field that would benefit from a hardware implementation and that you believe its a good research topic please share :)
 

Online blueskull

  • Supporter
  • ****
  • Posts: 11077
  • Country: cn
  • Power Electronics Guy
Re: Advice on PHD topic
« Reply #9 on: January 12, 2019, 10:45:11 am »
Thanks a lot!
So from these fields you suggest the computer vision one. What do you think about networking (schedulers for example)?
Also if you have any other field that would benefit from a hardware implementation and that you believe its a good research topic please share :)

Don't know much on networking, but hardware accelerated networking is definitely a topic of interest.
Intel used to make some network processors, so did (does?) Freescale.
But most network processors used nowadays are either implemented on FPGAs or build specifically for a product (by big gplayers, Cisco, Huawei, etc.).
Very few general purpose NPs are out there, and I think there could be a need on software defined network.

Google is trying to standardize their HTTP over UDP thing (Google QUIC), and I thin it can start a new paradigm on how people use network resource.
If TCP is to be phased out, a lot of things will happen, hence a lot of opportunities.
 
The following users thanked this post: trys11

Online Psi

  • Super Contributor
  • ***
  • Posts: 6810
  • Country: nz
Re: Advice on PHD topic
« Reply #10 on: January 12, 2019, 10:45:49 am »
He's an idea for consideration.
Wireless HD video systems for drone cameras exist but they are expensive $500-$2000, whereas you can buy hardware to do a standard definition video link over 1+km for $23 ($9 5.8ghz composite video receiver and $14 for the 600mW transmitter).

The cheap standard definition analog gear is totally uncompressed video over 5.8ghz. So the data link should be able to handle much better quality if the data had some compression with error checking and recovery.

So how about trying to design part of a system to do take HDMI video and compress/decompress it at ultra low latency.
It has to be ultra low latency because when you're flying a drone between trees at 70kph you just cant fly with lag.

Some features i think this system would need.
- Probably only frame compression, without temporal compression to keep latency ultra low.
- Compression doesn't need to be super high or complicated, just enough so it can be sent using cheap RF link hardware without any need for fancy Phase Shift Keying etc.
- Runs on cheap hardware/fpga
- Some sort of scaleable resolution decoding, So as you fly further away and the RF link has more errors you lose detail but still have enough to fly home. i.e. no hard wall for video reception where it just goes black.
This could be done using a side channel to return RSSI from receiver back to drone over the control link so it can reduce video data rate. Or it could be done automatically by the design of the video encoding system where data errors effect fine detail but not the overall picture. Ideally you'd want something similar to analog video, so as range increases the picture just gets grainy/B&W


The bonus would be that once you finish your PHD you can maybe make it into a product you can sell
« Last Edit: January 12, 2019, 11:52:01 pm by Psi »
Greek letter 'Psi' (not Pounds per Square Inch)
 
The following users thanked this post: trys11

Offline scatha

  • Regular Contributor
  • *
  • Posts: 53
  • Country: au
Re: Advice on PHD topic
« Reply #11 on: January 12, 2019, 12:05:02 pm »
The general advice given to PhD candidates at the moment is to avoid computer vision as a subject - the state-of-the-art performance is doubling each year, and each iteration introduces a new underlying architecture. This makes it pretty much impossible to begin a topic and have it even remotely relevant after three years - it's pretty dispiriting when this happens, believe me.

Hardware-accelerated networking is definitely a thing (particularly for the high-frequency trading community), but it's hard to think of an area which hasn't already been thoroughly explored. I mean, if you can come up with a novel scheduler which out-performs WFQ for a particular workload then great, but then most of your work is going to be done in network simulators, not in hardware. Software-defined networking I find a bit too high-level and buzzwordy.

There's currently a lot of work being put into hardware-accelerated machine learning, implementing feedforward/recurrent/convolutional neural networks on FPGAs, and developing novel architectures which map better to hardware (binarized neural networks, etc.). This might be worthwhile looking into, though it comes which similar caveats as CV.

Ultimately you're better off identifying potential supervisors and getting them to suggest topics.
 
The following users thanked this post: trys11

Offline trys11

  • Contributor
  • Posts: 11
  • Country: gr
Re: Advice on PHD topic
« Reply #12 on: January 12, 2019, 09:36:16 pm »
Thanks a lot!
So from these fields you suggest the computer vision one. What do you think about networking (schedulers for example)?
Also if you have any other field that would benefit from a hardware implementation and that you believe its a good research topic please share :)

Don't know much on networking, but hardware accelerated networking is definitely a topic of interest.
Intel used to make some network processors, so did (does?) Freescale.
But most network processors used nowadays are either implemented on FPGAs or build specifically for a product (by big gplayers, Cisco, Huawei, etc.).
Very few general purpose NPs are out there, and I think there could be a need on software defined network.

Google is trying to standardize their HTTP over UDP thing (Google QUIC), and I thin it can start a new paradigm on how people use network resource.
If TCP is to be phased out, a lot of things will happen, hence a lot of opportunities.

Thanks for your input
He's an idea for consideration.
Wireless HD video systems for drone cameras exist but they are expensive $500-$2000, whereas you can buy hardware to do a standard definition video link over 1+km for $23 ($9 5.8ghz composite video receiver and $14 for the 600mW transmitter).

The cheap standard definition analog gear is totally uncompressed video over 5.8ghz. So the data link should be able to handle much better quality if the data had some compression with error checking and recovery.

So how about trying to design part of a system to do take HDMI video and compress/decompress it at ultra low latency.
It has to be ultra low latency because when you're flying a drone between trees at 70kph you just cant fly with lag.

Some features i think this system would need.
- Probably only frame compression, without temporal compression to keep latency ultra low.
- Compression doesn't need to be super high or complicated, just enough so it can be sent using cheap RF link hardware without any need for fancy Phase Shift Keying etc.
- Runs on cheap hardware/fpga
- Some sort of scaleable resolution decoding, So as you fly further away and the RF link has more errors you lose detail but still have enough to fly home. i.e. no hard wall for video reception where it just goes black.
This could be done using a side channel to return RSSI from transmitter back to drone over the control link so it can reduce data rate. Or it could be done automatically by the design of the video encoding system where data errors effect fine detail but not the overall picture. Ideally you'd want something similar to analog video, so as range increases the picture just gets grainy/B&W


The bonus would be that once you finish your PHD you can maybe make it into a product you can sell
Thanks for your idea, it sounds good and it is in the spirit of making an implementation. I like it !
The general advice given to PhD candidates at the moment is to avoid computer vision as a subject - the state-of-the-art performance is doubling each year, and each iteration introduces a new underlying architecture. This makes it pretty much impossible to begin a topic and have it even remotely relevant after three years - it's pretty dispiriting when this happens, believe me.

Hardware-accelerated networking is definitely a thing (particularly for the high-frequency trading community), but it's hard to think of an area which hasn't already been thoroughly explored. I mean, if you can come up with a novel scheduler which out-performs WFQ for a particular workload then great, but then most of your work is going to be done in network simulators, not in hardware. Software-defined networking I find a bit too high-level and buzzwordy.

There's currently a lot of work being put into hardware-accelerated machine learning, implementing feedforward/recurrent/convolutional neural networks on FPGAs, and developing novel architectures which map better to hardware (binarized neural networks, etc.). This might be worthwhile looking into, though it comes which similar caveats as CV.

Ultimately you're better off identifying potential supervisors and getting them to suggest topics.
Thanks for your input as well! Cant a phd consist of a number of completed projects that were relevant at the time they were created?
Also, since I am looking for a funded phd program, I think I cant just find professors and discuss what I would like.
They come up with a funded project with a given description and you apply for it. That being said I think you will have limited freedom to do various things as long as it stays under the general title of the project (but this still is just my assumption).
 

Offline rhb

  • Super Contributor
  • ***
  • Posts: 2411
  • Country: us
Re: Advice on PHD topic
« Reply #13 on: January 13, 2019, 04:51:53 am »
A PhD dissertation must be original and successful.  That's the major distinction from the MS.

In reflection seismology getting a PhD at Stanford, UT Austin or Colorado School of Mines typically takes 4-6 years of work living on whatever the student stipend is.  It's much higher at Stanford than at Austin or Mines because of the cost of living in Palo Alto.  The supervisor is buying cheap (aka slave) labor.  But you go to the major  professional society meetings and the consortium meetings so you get to know all the important people in your field very well.

The BS is learning how to learn. The MS is learning something.  The PhD is learning how to teach yourself things no one knows.

You still have an MS level view which is clearly reflected in your questions and comments. The particular topic is going to be chosen by your supervisor with some leeway based on your abilities and interests.  But  your supervisor has to satisfy the interests of those providing the money.

I got a BA in English lit, an MS in igneous petrology and then spent 4 years at Austin pursuing a PhD in geophysics.  So no math to nothing but math.  A personal conflict with my supervisor led to the loss of financial support so I went back to work.  Going to Stanford or Mines would have been another 4-6 years of lost income and I could not justify it.

I got what the PhD is all about, being able to teach myself anything.  I just did not get my union card.  Except for not being able to get certain jobs, I effectively have a PhD.  And most people I've dealt with professionally have been surprised to learn I didn't get mine.  I also was compensated at similar rates as my friends who graduated from Stanford, Austin and Mines.

And people like Jim Williams who had no degrees at all blew most if not all PhD level analog EEs out of the water.

Any topic which is "hot" when you start, will not be when you finish.  So that's a very bad criterion.  The sole reason to do it is you want to acquire to ability to learn any topic at state of the art level.

I strongly urge you to look into "compressive sensing".  That's rather narrow relative to what the mathematics can do, so I call it "sparse L1 pursuits".  It's the most important development in applied mathematics since the work of Norbert Wiener during the 1940's and 1950's.  Wiener's work is the foundation of all digital signal processing up to the advent of wavelets.  Sparse L1 pursuits are wavelets on steroids.
 
The following users thanked this post: trys11

Offline RandallMcRee

  • Frequent Contributor
  • **
  • Posts: 295
  • Country: us
Re: Advice on PHD topic
« Reply #14 on: January 13, 2019, 05:00:56 am »
Looking at the list you provided in the original post...

I would think that there is an outstanding need in the world today for someone who deeply understands how to build wireless sensor networks that are completely, utterly, provably secure. Not only a PhD but a career (as Blueskull mentions this is a good thing).

Think the power grid, any modern factory and so on.

 
The following users thanked this post: trys11

Offline CatalinaWOW

  • Super Contributor
  • ***
  • Posts: 3013
  • Country: us
Re: Advice on PHD topic
« Reply #15 on: January 13, 2019, 05:53:31 am »
As has been hinted in a couple of these posts, the social aspect is nearly as important as the technical.  Your topic must interest you and an adviser.  An adviser who is interested in your topic must be willing and in a position to take on another student.  Your work must satisfy an adviser and committee.  Which includes being difficult enough that none of them would do it as a side topic.  They either thought the challenge was too great or they didn't have time to put into that area.

So spend time researching the staff at your university.   What their interests are.  What their reputation is.  How many students they already have and how that compares to their historical load.  Let that be part of what you use to direct your research.
 
The following users thanked this post: trys11

Offline trys11

  • Contributor
  • Posts: 11
  • Country: gr
Re: Advice on PHD topic
« Reply #16 on: January 13, 2019, 08:54:20 am »
A PhD dissertation must be original and successful.  That's the major distinction from the MS.

In reflection seismology getting a PhD at Stanford, UT Austin or Colorado School of Mines typically takes 4-6 years of work living on whatever the student stipend is.  It's much higher at Stanford than at Austin or Mines because of the cost of living in Palo Alto.  The supervisor is buying cheap (aka slave) labor.  But you go to the major  professional society meetings and the consortium meetings so you get to know all the important people in your field very well.

The BS is learning how to learn. The MS is learning something.  The PhD is learning how to teach yourself things no one knows.

You still have an MS level view which is clearly reflected in your questions and comments. The particular topic is going to be chosen by your supervisor with some leeway based on your abilities and interests.  But  your supervisor has to satisfy the interests of those providing the money.

I got a BA in English lit, an MS in igneous petrology and then spent 4 years at Austin pursuing a PhD in geophysics.  So no math to nothing but math.  A personal conflict with my supervisor led to the loss of financial support so I went back to work.  Going to Stanford or Mines would have been another 4-6 years of lost income and I could not justify it.

I got what the PhD is all about, being able to teach myself anything.  I just did not get my union card.  Except for not being able to get certain jobs, I effectively have a PhD.  And most people I've dealt with professionally have been surprised to learn I didn't get mine.  I also was compensated at similar rates as my friends who graduated from Stanford, Austin and Mines.

And people like Jim Williams who had no degrees at all blew most if not all PhD level analog EEs out of the water.

Any topic which is "hot" when you start, will not be when you finish.  So that's a very bad criterion.  The sole reason to do it is you want to acquire to ability to learn any topic at state of the art level.

I strongly urge you to look into "compressive sensing".  That's rather narrow relative to what the mathematics can do, so I call it "sparse L1 pursuits".  It's the most important development in applied mathematics since the work of Norbert Wiener during the 1940's and 1950's.  Wiener's work is the foundation of all digital signal processing up to the advent of wavelets.  Sparse L1 pursuits are wavelets on steroids.
I am sorry to hear you had such a bad experience, you didnt have the ability to change supervisor? I will surelly check  "compressive sensing" out, thanks !.
Looking at the list you provided in the original post...

I would think that there is an outstanding need in the world today for someone who deeply understands how to build wireless sensor networks that are completely, utterly, provably secure. Not only a PhD but a career (as Blueskull mentions this is a good thing).

Think the power grid, any modern factory and so on.


Thanks for your input!
As has been hinted in a couple of these posts, the social aspect is nearly as important as the technical.  Your topic must interest you and an adviser.  An adviser who is interested in your topic must be willing and in a position to take on another student.  Your work must satisfy an adviser and committee.  Which includes being difficult enough that none of them would do it as a side topic.  They either thought the challenge was too great or they didn't have time to put into that area.

So spend time researching the staff at your university.   What their interests are.  What their reputation is.  How many students they already have and how that compares to their historical load.  Let that be part of what you use to direct your research.
Thanks.
Since I am/will be applying in already funded projects I believe that they will all be in the field of interest of any given professor.
About reputation, is it counted by the number of citations? Also how much does the university ranking count?
 

Online emece67

  • Regular Contributor
  • *
  • Posts: 111
  • Country: es
Re: Advice on PHD topic
« Reply #17 on: January 13, 2019, 09:57:38 am »
Hi,

Processing of hyperspectral images has applications in many fields, each of them using not exactly the same techniques and, sometimes, using completely different ones. Being you a hardware guy with previous experience on CV maybe you can build some system suited for a particular task.

I'm not at all an expert, neither I'm working on this field, but some of my colleagues are (they are working on brain surgery & remote sensing for crop science ---not the journal---) and, well, they have a lot of work to do and companies willing to put money on this.

Hope this helps, regards.
Information must flow.
 
The following users thanked this post: trys11

Offline rhb

  • Super Contributor
  • ***
  • Posts: 2411
  • Country: us
Re: Advice on PHD topic
« Reply #18 on: January 13, 2019, 12:45:21 pm »
I was warned I would not get my degree under my supervisor my first year by some friends who were academics.  I ignored the advice.  I don't recall what I told them, but they immediately got very concerned.  But I didn't want to go to Stanford because of the population and earthquake density.

Citations and school ranking are good metrics to start with, but interpersonal relationship is more important.  Read prospective supervisors papers, establish communication with them.  Had I done that I'd have gone to Stanford instead and not bothered to apply to Austin.  I made two applications for PhD programs, both at Austin.  One in EE and the other in geophysics.  Geophysics offered me  a research assistantship.  I paid my own way through my MS.  I was not going to do that for a PhD.

My supervisor was one of Norbert Wiener's students in the Geophysical Analysis Group, had been head of research at a major seismic service company.  But all he was actually known for was using a magnetic drum to implement deconvolution of water bottom reflections on analog recordings in the late 50's.

BTW the best place to start with compressive sensing is the discussion of basis pursuit in the 3rd ed of Mallat's "A Wavelet Tour of Signal Processing".

I urge you to get the following books:

A Wavelet Tour of Signal Processing
Stephane Mallat
3rd ed

Random Data
Bendat and Piersol
4th ed

A Mathematical Introduction to Compressive Sensing
Foucart and Rauhut

F&R is the most difficult book I ever read.  Mallat is in 2nd place.  By the time I found B&P I knew all the material already, so I have just used it as a reference starting with the 2nd ed.

An important thing to understand is that while the mathematical proof of compressive sensing is *really* difficult, actual practice is as easy as doing an FFT.  Compressive sensing was developed by David Donoho.  One of his proofs is 15 pages long!  After completing that he commented that the reader will doubtless be relieved that theorems 2 & 3 are much simpler and took only 3-4 lines each.

Go here:

https://statistics.stanford.edu/resources/technical-reports

and read Donoho's papers from 2004.  Emmanuel Candes reports from that period are here:

https://statweb.stanford.edu/~candes/publications.html

Both are very good writers, so focus on the introductions and leave the proofs for later.  Having been rigorously schooled in Wiener's work, Donoho and Candes blow my mind even after 5 years and 3000 pages.  Donoho proved you could do what I had been taught was impossible in this paper:

https://statistics.stanford.edu/research/most-large-underdetermined-systems-linear-equations-minimal-l1-norm-solution-also-sparsest
 
The following users thanked this post: trys11

Offline tomato

  • Regular Contributor
  • *
  • Posts: 154
  • Country: us
Re: Advice on PHD topic
« Reply #19 on: January 13, 2019, 01:37:52 pm »

Also, any general tips in starting a phd would be MUCH appreciated.

Don't put too much pressure on yourself trying to pin down your research topic so early in your schooling.  Some PhD students have a topic their first year, but most do not.  Many do not even know who their thesis advisor will be until after one or two semesters.  Be patient, and it will all fall into place.
 
The following users thanked this post: trys11

Offline trys11

  • Contributor
  • Posts: 11
  • Country: gr
Re: Advice on PHD topic
« Reply #20 on: January 14, 2019, 04:02:23 am »
Hi,

Processing of hyperspectral images has applications in many fields, each of them using not exactly the same techniques and, sometimes, using completely different ones. Being you a hardware guy with previous experience on CV maybe you can build some system suited for a particular task.

I'm not at all an expert, neither I'm working on this field, but some of my colleagues are (they are working on brain surgery & remote sensing for crop science ---not the journal---) and, well, they have a lot of work to do and companies willing to put money on this.

Hope this helps, regards.
It does help, thanks a lot :)
I was warned I would not get my degree under my supervisor my first year by some friends who were academics.  I ignored the advice.  I don't recall what I told them, but they immediately got very concerned.  But I didn't want to go to Stanford because of the population and earthquake density.

Citations and school ranking are good metrics to start with, but interpersonal relationship is more important.  Read prospective supervisors papers, establish communication with them.  Had I done that I'd have gone to Stanford instead and not bothered to apply to Austin.  I made two applications for PhD programs, both at Austin.  One in EE and the other in geophysics.  Geophysics offered me  a research assistantship.  I paid my own way through my MS.  I was not going to do that for a PhD.

My supervisor was one of Norbert Wiener's students in the Geophysical Analysis Group, had been head of research at a major seismic service company.  But all he was actually known for was using a magnetic drum to implement deconvolution of water bottom reflections on analog recordings in the late 50's.

BTW the best place to start with compressive sensing is the discussion of basis pursuit in the 3rd ed of Mallat's "A Wavelet Tour of Signal Processing".

I urge you to get the following books:

A Wavelet Tour of Signal Processing
Stephane Mallat
3rd ed

Random Data
Bendat and Piersol
4th ed

A Mathematical Introduction to Compressive Sensing
Foucart and Rauhut

F&R is the most difficult book I ever read.  Mallat is in 2nd place.  By the time I found B&P I knew all the material already, so I have just used it as a reference starting with the 2nd ed.

An important thing to understand is that while the mathematical proof of compressive sensing is *really* difficult, actual practice is as easy as doing an FFT.  Compressive sensing was developed by David Donoho.  One of his proofs is 15 pages long!  After completing that he commented that the reader will doubtless be relieved that theorems 2 & 3 are much simpler and took only 3-4 lines each.

Go here:

https://statistics.stanford.edu/resources/technical-reports

and read Donoho's papers from 2004.  Emmanuel Candes reports from that period are here:

https://statweb.stanford.edu/~candes/publications.html

Both are very good writers, so focus on the introductions and leave the proofs for later.  Having been rigorously schooled in Wiener's work, Donoho and Candes blow my mind even after 5 years and 3000 pages.  Donoho proved you could do what I had been taught was impossible in this paper:

https://statistics.stanford.edu/research/most-large-underdetermined-systems-linear-equations-minimal-l1-norm-solution-also-sparsest
Thanks a lot for sharing all this information !

Also, any general tips in starting a phd would be MUCH appreciated.

Don't put too much pressure on yourself trying to pin down your research topic so early in your schooling.  Some PhD students have a topic their first year, but most do not.  Many do not even know who their thesis advisor will be until after one or two semesters.  Be patient, and it will all fall into place.
Thanks for the tip !:D
 

Offline jfiresto

  • Regular Contributor
  • *
  • Posts: 76
  • Country: de
Re: Advice on PHD topic
« Reply #21 on: January 14, 2019, 04:50:08 am »
... I got what the PhD is all about, being able to teach myself anything....

That is what many of us learned as undergraduates.

The point of a PhD is to demonstrate that you can find and define a tractable problem, become the world's expert on it, push to its solution (or suggest something more fruitful) and then teach the world your original work so as to advance your field. Just a tip: You don't actually have to solve your PhD problem. If you don't, your final defense will likely prove much easier. Very few people are able to criticize an unsuccessfully result as just the opposite. Ask me how I know.  :)

If I can give you one piece of advice it would be: spare no effort in finding a good PhD advisor! Your choice of advisor will likely decide whether your next 4+ years will be more pain or more pleasure, or ultimately, a success or a failure.
 
The following users thanked this post: trys11

Offline trys11

  • Contributor
  • Posts: 11
  • Country: gr
Re: Advice on PHD topic
« Reply #22 on: January 14, 2019, 05:04:33 am »
... I got what the PhD is all about, being able to teach myself anything....

That is what many of us learned as undergraduates.

The point of a PhD is to demonstrate that you can find and define a tractable problem, become the world's expert on it, push to its solution (or suggest something more fruitful) and then teach the world your original work so as to advance your field. Just a tip: You don't actually have to solve your PhD problem. If you don't, your final defense will likely prove much easier. Very few people are able to criticize an unsuccessfully result as just the opposite. Ask me how I know.  :)

If I can give you one piece of advice it would be: spare no effort in finding a good PhD advisor! Your choice of advisor will likely decide whether your next 4+ years will be more pain or more pleasure, or ultimately, a success or a failure.
Thanks for your advice. I have already felt how important it is to have good communication/relationship with both your colleagues and professors but the thing is that i will apply for a position in a University I have never visited and to a professor, I have never known. Also making a phd in the University I have graduated from is out of the question since the funding is abysmal and so is in most greek universities that I know of.
 

Offline jfiresto

  • Regular Contributor
  • *
  • Posts: 76
  • Country: de
Re: Advice on PHD topic
« Reply #23 on: January 14, 2019, 05:39:43 am »
... the thing is that i will apply for a position in a University I have never visited and to a professor, I have never known.

Unless you want play a form of Russian roulette, I would find some way to visit and interview your potential professor(s). Find out if you are compatible. Interview his students. See what support you will have. See what support your Professor has. Find out how many years his students take to finish, and what jobs or positions they found afterwards, or their professor found for them. Talk to those former students.

Quote
Also making a phd in the University I have graduated from is out of the question since the funding is abysmal and so is in most greek universities that I know of....

My undergraduate university forbid it, as they consider it academic incest. Which it is.
 
The following users thanked this post: trys11

Offline rhb

  • Super Contributor
  • ***
  • Posts: 2411
  • Country: us
Re: Advice on PHD topic
« Reply #24 on: January 14, 2019, 06:22:36 am »
Do not make the mistake I did.  Go spend several *days*  talking to your prospective professor and other members of the department, students and staff.  Read the professors papers and especially any consortium reports he has written.  Published papers will have been reviewed and edited.  Consortium reports will not.  Had I done that I'd have gone to Stanford instead and would have gotten my degree.

I got an immediate offer on my one day (a few hours really) visit to Austin and went off with a box of consortium reports.  In reading them I found that there were long sections that were absolutely awful writing.  I got a sick feeling when I realized they were my supervisor's contributions.

A PhD will cost you well over $100K in lost income.  Probably several times that.  Spend that money very carefully.

@ jfiresto  I taught myself algebra at age 12 and trigonometry at 14.  At university I took a BA in English literature for which I read as many books by the authors we were reading on the side as I read for my courses and then, after a couple of years as a cafeteria manager,  an MS in igneous petrology.  For lack of employment in geology I went into geophysics.  I liked it so I went back to school to better learn the mathematics as I had stopped at differential equations. And I had *no* formal education in seismology at all.

I terminated my contract with a super major in 2007 to move to Arkansas to look after my aging parents.  I'd expected to get offsite contract work, but the 2008 crash prevented that from happening.

In 2013 I ran into the work of Emmanuel Candes and David Donoho though I did not know at the time that they were the original authors.  Fortuitously, "A Mathematical Introduction to Compressive Sensing" by Foucart and Rauhut appeared.  It took two readings with Mallat's 3rd ed in between and a very large number of papers by Donoho, Candes and their students to understand how and why it worked.  F&R is 600 pages of pure mathematics.  I invite you to get a copy and let us know if you learned to teach yourself that as an undergraduate in any field outside of mathematics program.

I can't say what is allowed for a PhD dissertation topic these days.  Standards are so low I'd believe almost anything.   But in my MS professor's day, a negative result on a PhD project meant starting over.  And having attended many Stanford, Austin and Mines consortia meetings I *very* much doubt they allow a negative result on a dissertation topic.  So the students spend 2-3 years finding a project which will succeed.  I have *never* seen a negative result in a dissertation from any of those schools.  And I had an almost complete set of Stanford Exploration Project theses and dissertations in my office for many years along with a pretty large part of the work done at Mines.

The undergraudate level is just learning how to learn.  The MS you learn something and the PhD is supposed to enable you to advance you past all barriers.

BTW My supervisor granted a PhD to a student who did not know Snell's law a couple of years into the PhD program.  By that time my supervisor was also completely blind from retinal detachments.

Be very wary of anyone who advertises they did not get their PhD.
 
The following users thanked this post: trys11


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf