EEVblog Electronics Community Forum
General => General Technical Chat => Topic started by: MathWizard on July 28, 2024, 01:02:25 am
-
I guess IDK what anyone has to do for a PhD, and surely most people aren't discovering new things to research.
So besides doing a certain amount of courses, etc, what sort of thesis would someone do to get an EE PhD degree. I think I've heard of a master's thesis too, so again what type's of thing do people do ?
Some day I might go back to my local university or do some online EE courses, but no way do I want to do a full time course load, or get more than a bachelor's, any time soon. Plus it costs a fortune.
-
It's all applied stuff. The PhDs I work with are very deep down the signal processing rabbit hole. They're not researching a new and improved FFT or optimizer, for example, but might be figuring out how to apply an FFT or optimizer to solve a a real-world problem.
Though the few PhD defenses I've recently attended are all of the form "we used a neural network and magically solved all our problems"... ::)
-
In power electronics, there's lots of people doing algorithm optimisation. Usually a shed load of maths followed by a lot of simulation with too many significant figures (efficiency 99.4553%, inductance 15.656uH etc). Having done all that they spend a few months getting it to work on a piece of hardware developed a few years ago by someone else and claim great success as the efficiency improves from 65% to 80% while everyone in the market has 96%+
Maybe I'm looking at the wrong ones...
There are also people doing wide bandgap device research. Eg developing novel SiC MOSFETS and GaN HEMTs. There's a lot of time spent in cleanrooms for that sort of PhD and blowing up the devices you've just made.
Also yes, magic neural nets do often turn up. Saw one in a forensics PhD recently. It did not solve all the problems.
-
Currently, I think the most interesting PhDs would be in microelectronics, where there's still a lot of interesting stuff to explore.
-
Lots of effort by PhDs and Post Docs goes into developing the leading Semiconductor technology, since this is the realm where conventional physics says "You Can't Do That!!" :o
What's absolutely amazing is how quick this SOTA Semiconductor Technology gets to the public, just look at the latest Apple Chips fabbed by TSMC at 3nm!!
Most PhDs want to push the SOTA envelope or teach at the university level.
What many folks don't realize is a PhD candidate must possess a broad fundamental knowledge base in the field to acquire such, otherwise they wouldn't have been credited with a PhD from a reputable university!
Wether the PhD retains this broad fundamental knowledge is another story ???
Best,
-
While in HP Labs half a lifetime ago, a group of us pondered "why do a PhD?".
To make yourself more employable? Not really, since if you could do a PhD you were pretty employable anyway, and your PhD's subject is unlikely to be of direct interest to most employers. Plus, once they've sucked your experience out of you, they may decide they don't need an expensive specialist anymore.
To make more money? Not really; that would require directly relevant experience.
To be able to climb further up the greasy pole? Definitely necessary in an academic career, might still be necessary in some government roles.
Because you want to do it? An excellent reason, possibly the only excellent reason.
Those that hope to be able to finish/write their PhD while working in a job are probably underestimating the difficulty :)
EDIT: add "to be able to migrate to another country" is possibly a good reason, since it is a concrete demonstration that you have specialist skills. But the skills do have to be in demand, and the PhD has to be from an institution with an international (good!) reputation.
-
When people like AT&T, HP and IBM has substantial world famous research labs, quite a few people got a PhD in an engineering discipline and moved to those places. These days I guess they either stay in academia, or compete in the job market against someone with a masters degree and more industrial experience.
There is still plenty of R&D to do in both the core hardware (e.g. semiconductors) and applications (e.g. comms) areas of electronics, so its not hard to pick a relevant area in which to do new work.
-
While in HP Labs half a lifetime ago, a group of us pondered "why do a PhD?".
To make yourself more employable? Not really, since if you could do a PhD you were pretty employable anyway, and your PhD's subject is unlikely to be of direct interest to most employers. Plus, once they've sucked your experience out of you, they may decide they don't need an expensive specialist anymore.
To make more money? Not really; that would require directly relevant experience.
To be able to climb further up the greasy pole? Definitely necessary in an academic career, might still be necessary in some government roles.
Because you want to do it? An excellent reason, possibly the only excellent reason.
Those that hope to be able to finish/write their PhD while working in a job are probably underestimating the difficulty :)
When I graduated in the 70s the commonest reason for someone to go on to a PhD program was they couldn't find a job. From some things I've heard about the bad job market this year, the same thing seems to apply today.
-
While in HP Labs half a lifetime ago, a group of us pondered "why do a PhD?".
To make yourself more employable? Not really, since if you could do a PhD you were pretty employable anyway, and your PhD's subject is unlikely to be of direct interest to most employers. Plus, once they've sucked your experience out of you, they may decide they don't need an expensive specialist anymore.
To make more money? Not really; that would require directly relevant experience.
To be able to climb further up the greasy pole? Definitely necessary in an academic career, might still be necessary in some government roles.
Because you want to do it? An excellent reason, possibly the only excellent reason.
Those that hope to be able to finish/write their PhD while working in a job are probably underestimating the difficulty :)
When I graduated in the 70s the commonest reason for someone to go on to a PhD program was they couldn't find a job. From some things I've heard about the bad job market this year, the same thing seems to apply today.
I could believe that if the PhD is, in effect, a means to delay making a choice. I doubt it contradicted my points above.
Before I graduated in '78, I made (IIRC!) 11 job applications, was offered interviews at ~16 companies(!), and received a couple of letters that I interpreted as "we are interviewing and you can have a job". I didn't bother to go to such interviews!
-
Before I graduated in '78, I made (IIRC!) 11 job applications, was offered interviews at ~16 companies(!), and received a couple of letters that I interpreted as "we are interviewing and you can have a job". I didn't bother to go to such interviews!
Why didn't you go those interviews? Some of them are just a way to gather a bunch of deadbeats to charge to the government. Others are people building a substantial team, for a large new activity, where they almost certainly have a job for you if you have a degree from a respectable university. They only question in where in the pecking order they might put you. Those can be good opportunities to rise quickly, because there are constantly openings in a place like that.
-
Before I graduated in '78, I made (IIRC!) 11 job applications, was offered interviews at ~16 companies(!), and received a couple of letters that I interpreted as "we are interviewing and you can have a job". I didn't bother to go to such interviews!
Why didn't you go those interviews? Some of them are just a way to gather a bunch of deadbeats to charge to the government. Others are people building a substantial team, for a large new activity, where they almost certainly have a job for you if you have a degree from a respectable university. They only question in where in the pecking order they might put you. Those can be good opportunities to rise quickly, because there are constantly openings in a place like that.
They were at GEC/Marconi sites that wanted as many bums on seats as they could get for, as I later discovered, government cost-plus contracts.
I rapidly determined I never wanted to work for that type of place. That was an easy decision after an experience at one site. After a days interview I was chatting to the gate security guards, and they asked what I was doing. I muttered about the milk round and seeing what places I would like to work. The guard commented "And I'm sure there will be other places". I took the hint and never regretted it!
-
Before I graduated in '78, I made (IIRC!) 11 job applications, was offered interviews at ~16 companies(!), and received a couple of letters that I interpreted as "we are interviewing and you can have a job". I didn't bother to go to such interviews!
Why didn't you go those interviews? Some of them are just a way to gather a bunch of deadbeats to charge to the government. Others are people building a substantial team, for a large new activity, where they almost certainly have a job for you if you have a degree from a respectable university. They only question in where in the pecking order they might put you. Those can be good opportunities to rise quickly, because there are constantly openings in a place like that.
They were at GEC/Marconi sites that wanted as many bums on seats as they could get for, as I later discovered, government cost-plus contracts.
I rapidly determined I never wanted to work for that type of place. That was an easy decision after an experience at one site. After a days interview I was chatting to the gate security guards, and they asked what I was doing. I muttered about the milk round and seeing what places I would like to work. The guard commented "And I'm sure there will be other places". I took the hint and never regretted it!
1978 was kind of a special year. There were lots of jobs around that year. People were going to Australia and other countries, trying to recruit their fresh and recent engineering grads to work in the UK. A year or two before and a couple of years later you might have been glad of one of the GEC jobs to get you started. :)
-
EE PhD in Canada? Extremely rare unless you want to become a prof - it contrasts with engineering as it is an applied science.
We don't have a decent technology industry here that can make use of them.
I see many thesis from developing, third-world countries that are basically garbage, just a repost of an IC datasheet. Amazing how little is needed to get one there.
If you can narrow it (thesis) down - to something "popular" where the grants and funding are, your Supervisor's lab interests, that makes it much easier than a thesis about something out on left field.
The only PhD's in engineering I've known were the guys who graduated and couldn't find a job. So they went back and got a Masters. After that couldn't find a job, so they got a PhD. In the end way too expensive and the massive debt, and then reality creeps in.
One EE PhD I know he clings to any job he has because there are no EE PhD job postings out West. They are viewed as too expensive, too specialized so employability is not there. Ontario has some postings on Indeed etc. I'm saying also look ahead at where the PhD will take you.
If I was doing it, you should innovate, come up with something better, solve a problem instead of looking to do a thesis on a topic others give.
-
I finished my PhD last year. Some of my papers are public access, all are linked from my google scholar profile: https://scholar.google.com/citations?user=H9L4UYcAAAAJ&hl=en
My research topic was on using piezoelectric resonators to replace inductors for higher power density in dc/dc converters. The work consisted of modeling and fabricating piezoelectric devices in the cleanroom and modeling and building dc/dc converters that used them. When I proposed the project to my advisor there were only a few recent papers on the topic. The research area has since received more focus and a number of research groups at different universities are working on it. It still remains to be seen if it will be a viable technology, there are a number of fundamental challenges to be overcome. I got to come up with the project idea myself and by the time I graduated the project was funded with a few students working on it and we had a collaborator at a different university, so it was cool to watch it all develop.
To receive a PhD you need to make a novel contribution to the field, which is one of the things separating a masters degree from a PhD. However, I would say that the novel contribution can sometimes be pretty contrived.
In engineering, PhD programs play a big role in job training for some subfields. For IC design, and especially analog or mixed signal, you basically need a PhD to get hired. When I was at Tesla, most of the power electronics team had PhDs, but it seemed a bit overkill to me.
This is US centric advice, but if you are a US permanent resident the downsides of getting a PhD in terms of opportunity cost outweigh the benefits unless you are entering a top ranked program. If you are an international student looking to move to the USA any program might be worth your time. If your field is in demand you can get an O-1 visa, which is not capped or lottery based like the H-1B visa. I am not familiar with the process but I think there are also some other visa opportunities that benefit PhD holders.
In no case should anyone be paying tuition for a PhD in STEM. The stipend that is provided is not great but should be enough to live on.
-
If you are an international student looking to move to the USA any program might be worth your time. If your field is in demand you can get an O-1 visa, which is not capped or lottery based like the H-1B visa. I am not familiar with the process but I think there are also some other visa opportunities that benefit PhD holders.
Valid point, added to my previous list.
The PhD should be from an internationally recognised institution with a good reputation, of course. Predicting which skill will be in demand in several years time after is PhD is, as they say, left an an exercise for the student.
-
My research topic was on using piezoelectric resonators to replace inductors for higher power density in dc/dc converters. The work consisted of modeling and fabricating piezoelectric devices in the cleanroom and modeling and building dc/dc converters that used them. When I proposed the project to my advisor there were only a few recent papers on the topic. The research area has since received more focus and a number of research groups at different universities are working on it. It still remains to be seen if it will be a viable technology, there are a number of fundamental challenges to be overcome. I got to come up with the project idea myself and by the time I graduated the project was funded with a few students working on it and we had a collaborator at a different university, so it was cool to watch it all develop.
Was that in any way related to the upsurge in interest in recent years for piezo approaches to converting mechanical energy to electrical energy in energy harvesting applications?
In engineering, PhD programs play a big role in job training for some subfields. For IC design, and especially analog or mixed signal, you basically need a PhD to get hired. When I was at Tesla, most of the power electronics team had PhDs, but it seemed a bit overkill to me.
That's an exaggeration. You need a masters to be considered for any decent engineering job these days. A bachelors no longer cuts it. However, I don't know any areas where a PhD is the norm. I've known very few IC designers who have one.
-
College courses often start teaching applications of physics, applications of math and applications with already invented circuits. We call this engineering, with the main goal of understanding and applying circuits. However many engineers will use standard building blocks, like the standard opamps circuits and off the shelf DC/DCs, and then call it a day.
A masters can go into more in depth theories or generalizations, with more comprehensive evaluations of circuits. Students will learn to read and derive new models and apply them. An average master thesis will include a literature study of state of the art research and then doing something with it. That can range from applying it or taking a stab at extending or doing something different. For many students this can be a bit of a trial to see if perhaps a PhD will fit them.. A lot of master student projects can eventually be published by either a supervisor or the student itself.
If a student likes doing research, writing it properly and making sure it collaborates within the community of research (instead of just being a tone-deaf plea), then go ahead and start a PhD.
There are plenty of topics to do a PhD in EE. Circuit design is a prime example, especially for chip design. Electrification of all industries is still not complete, so power electronics is still necessary. We also need build better amplifiers, RF mixers, frequency synthesizers and receivers for next generation modems. 5G or 6G doesn't come falling from the sky for free, and it's a collaboration between industry and academia to derive a new standard. Same goes with better ADCs, DACs, PLLs, etc. And antenna's are also designed by EEs, with lots of new stuff tried as well (new frequencies, shapes, arrays, multi-modes, etc.).
We need people to design PDKs for every new CMOS process node. EE can span well into computer engineering and embedded systems in order to understand what those sectors need to succeed well. Likewise, there is a lot of applications in signal processing, where linear system theory, probability theory and electronics is necessary to work on new DSP algorithms. This can span from a highly specialized algorithm only used by 1 application to something that is possibly generalized like approximate computing. Most research projects however do need a vehicle of research, so sometimes applications are a bit of a side effect to what the research is really about.
There is also lots of research on mixed signal designs (like partial analog, partial digital computing), or completely new architectures. System design is a giant topic in itself. Some PhDs will be about designing a single component in a new or better way, whereas other PhDs want to demonstrate novel applications by combining state-of-the-art research and extending upon it.
-
Ok some interesting answers, I guess the school you attend will play a big part in what you can research or do applications for.
I live on the coast, so marine related industry/research would be the most common stuff. So yeah there's lots of stuff getting modernized. I'd say there's some fun EE jobs where I live, for anyone getting out of school. If I ever did get a real EE job, I'm too old to be going out to sea though.
-
A young brilliant Cornell PhD Student's dissertation, Angle Sensitive Pixels for Integrated Light Field Sensing (also a few IEEE papers on such), was a new type CMOS Imaging chip based upon capturing photon intensity and angle of arrival at the pixel level which allows post image capture focus.
This and the IEEE papers gathered the attention of Apple, which subsequently offered a "no strings attached" extremely large sum of stock for an interview. Not a sign on bonus, or stock option, just come out for an interview, after that it's all yours whatever you decide!! Apple was successful in luring this new PhD for the interview and hire. Likely one can "see" the effects of his work in how iPhone cameras operate today.
Another brilliant Cornell PhD Student dissertation, "Analysis and Design of Wideband Passive Mixer-First Receivers (also authored a number of IEEE papers) based upon the Mixer-First or PolyPhase Mixer topology graduated and joined a startup call Passiff. The name hints at what they were working on, Passive Mixers!! A year later Apple acquired Passiff and she became an extremely wealthy young PhD!! Her work is likely in todays iPods, iPhones and iPads.
These students were products of Dr Al Molnar's Group at Cornell. Al's Groups are some of the brightest students we have ever encountered/worked with, and certainly hope this continues, altho we are out-of-the-loop now :(
https://molnargroup.ece.cornell.edu/people/
Anyway, a PhD can be rewarding both intellectually and financially if the right topic, advisor, university and more importantly the right person is behind such :-+
Best,
-
A young brilliant Cornell PhD Student's dissertation, Angle Sensitive Pixels for Integrated Light Field Sensing (also a few IEEE papers on such), was a new type CMOS Imaging chip based upon capturing photon intensity and angle of arrival at the pixel level which allows post image capture focus.
This and the IEEE papers gathered the attention of Apple, which subsequently offered a "no strings attached" extremely large sum of stock for an interview. Not a sign on bonus, or stock option, just come out for an interview, after that it's all yours whatever you decide!! Apple was successful in luring this new PhD for the interview and hire. Likely one can "see" the effects of his work in how iPhone cameras operate today.
Another brilliant Cornell PhD Student dissertation, "Analysis and Design of Wideband Passive Mixer-First Receivers (also authored a number of IEEE papers) based upon the Mixer-First or PolyPhase Mixer topology graduated and joined a startup call Passiff. The name hints at what they were working on, Passive Mixers!! A year later Apple acquired Passiff and she became an extremely wealthy young PhD!! Her work is likely in todays iPods, iPhones and iPads.
These students were products of Dr Al Molnar's Group at Cornell. Al's Groups are some of the brightest students we have ever encountered/worked with, and certainly hope this continues, altho we are out-of-the-loop now :(
https://molnargroup.ece.cornell.edu/people/
Anyway, a PhD can be rewarding both intellectually and financially if the right topic, advisor, university and more importantly the right person is behind such :-+
Best,
You don't need PhD people for that. Shannon laid the groundwork for the digital electronics industry in his master's dissertation. :)
More seriously. If you spend 3 or 4 years to complete a PhD the world can look very different at the end. I remember someone doing a PhD in positional sensing techniques and as he graduated the construction industry was facing new regulations for safety systems, like tipping sensors on cranes. These two things aligned beautifully, and he seemed to get rapidly wealthy from his PhD. However, that time alignment seemed like nothing more than luck. Luck that he had the get up and go to take good advantage of, but luck nevertheless.
-
You don't need PhD people for that. Shannon laid the groundwork for the digital electronics industry in his master's dissertation. :)
Andrew Viterbi did quite well with his PhD dissertation at USC, his BS and MS were from MIT where he studied under Wiener and Shannon, thus the foundation was laid!!
Best,
-
Some day I might go back to my local university or do some online EE courses, but no way do I want to do a full time course load, or get more than a bachelor's, any time soon. Plus it costs a fortune.
If you're paying for a PhD, you're doing it wrong. Either your employer should pay, or you should be getting paid as a research assistant.
-
Some day I might go back to my local university or do some online EE courses, but no way do I want to do a full time course load, or get more than a bachelor's, any time soon. Plus it costs a fortune.
If you're paying for a PhD, you're doing it wrong. Either your employer should pay, or you should be getting paid as a research assistant.
Yep. But that's going to be tough for someone older than 35-40, which I assumed was the case here (unless indeed their current employer is willing to, but that's also not ultra-likely).
-
Though the few PhD defenses I've recently attended are all of the form "we used a neural network and magically solved all our problems"... ::)
I am not in any academic field at all, but this sentence makes me curious: Are defenses like that actually successfull?
-
Though the few PhD defenses I've recently attended are all of the form "we used a neural network and magically solved all our problems"... ::)
... to which the next questions are "how can you prove the problem was completely solved?" followed by "where is the boundary where the problem was no longer completely solved?".
ML inherently has problems explaining why, and there are numerous examples where a tiny change caused failure.
-
Though the few PhD defenses I've recently attended are all of the form "we used a neural network and magically solved all our problems"... ::)
... to which the next questions are "how can you prove the problem was completely solved?" followed by "where is the boundary where the problem was no longer completely solved?".
ML inherently has problems explaining why, and there are numerous examples where a tiny change caused failure.
Well, that's the same for human learning. Very few things actually get rigorously analyzed, whether humans or machines are involved. We rely heavily on some pretty woolly assumptions about the universality of things in many areas. In fact, there are many adaptive things where its obvious they won't always converge properly, but do so a high enough percentage of the time to be very useful.
-
Some day I might go back to my local university or do some online EE courses, but no way do I want to do a full time course load, or get more than a bachelor's, any time soon. Plus it costs a fortune.
If you're paying for a PhD, you're doing it wrong. Either your employer should pay, or you should be getting paid as a research assistant.
One of our former colleagues went to Vanderbilt University in Nashville, TN for his PhD. The company in Largo, FL paid for everything including an apartment, travel, even his salary. This was in the early days of DSP for Communications (Modems) and the company became a world leader in such due to his work. Of course he had to agree to stay on for a number of years for the PhD benefit!!
As we stated earlier, "a PhD can be rewarding both intellectually and financially if the right topic, advisor, university and more importantly the right person is behind such"
Best,
-
Though the few PhD defenses I've recently attended are all of the form "we used a neural network and magically solved all our problems"... ::)
... to which the next questions are "how can you prove the problem was completely solved?" followed by "where is the boundary where the problem was no longer completely solved?".
ML inherently has problems explaining why, and there are numerous examples where a tiny change caused failure.
Well, that's the same for human learning. Very few things actually get rigorously analyzed, whether humans or machines are involved. We rely heavily on some pretty woolly assumptions about the universality of things in many areas. In fact, there are many adaptive things where its obvious they won't always converge properly, but do so a high enough percentage of the time to be very useful.
For everyday autonomic actions etc, yes that is how humans tend to behave. And all you have to do is look at people's behaviour and/or "reality TV" programmes to see how grossly deficient that is. The word "idiocracy" springs to mind.
Unfortunately ML neural nets are descendents of Igor Alexander's WISARD. That distinguished well between cars and tanks in the lab, but failed dismally in the field. Eventually they realised it had trained itself to distinguish between cloudy and sunny days. It is said colleagues then refused to acknowledge Alexander's presence on sunny days :)
There are documented examples where changing one pixel in a photo of a road "stop" sign, caused the ML system to change to interpreting it as a "40MPH" sign.
ML radiography interpreters was found to be deciding whether or not a cancer was treatable based on the font used in the radiograph. The training set correctly detected that one font was associated with a hospital in a poor area where the treatment outcomes were poorer.
ML in the (US) judicial system decides that black people shouldn't be released on bail, since that is the prejudice embodied in the training set.
Then there are the more subtle legal consequences - if a company using ML for employment decisions is accused of racial bias, they won't be able to disprove it.
FFI and references, see comp.risks; the 40 year archive is at http://catless.ncl.ac.uk/Risks/ (http://catless.ncl.ac.uk/Risks/)
-
Unfortunately ML neural nets are descendents of Igor Alexander's WISARD. That distinguished well between cars and tanks in the lab, but failed dismally in the field. Eventually they realised it had trained itself to distinguish between cloudy and sunny days. It is said colleagues then refused to acknowledge Alexander's presence on sunny days :)
Quoting a famous incompetently engineered example doesn't really say anything useful. Those people were arrogant idiots. Imperial seems to produce more than its fair share of those.
There are documented examples where changing one pixel in a photo of a road "stop" sign, caused the ML system to change to interpreting it as a "40MPH" sign.
This is an issue with current cars. Sometimes a very dirty and unclear sign fools them, and you can understand it. Other times you can drive through an area of road works with numerous speed restriction signs, all nice and clean and clear, and the car misreads every one in the same odd way. As of this month cars sold in the UK are required to have a road sign reader nagging the driver. Some wanted that to enforce the speed restriction, which thankfully didn't get through.
ML radiography interpreters was found to be deciding whether or not a cancer was treatable based on the font used in the radiograph. The training set correctly detected that one font was associated with a hospital in a poor area where the treatment outcomes were poorer.
ML in the (US) judicial system decides that black people shouldn't be released on bail, since that is the prejudice embodied in the training set.
Then there are the more subtle legal consequences - if a company using ML for employment decisions is accused of racial bias, they won't be able to disprove it.
FFI and references, see comp.risks; the 40 year archive is at http://catless.ncl.ac.uk/Risks/ (http://catless.ncl.ac.uk/Risks/)
There are a lot of poor quality engineers out there. They'll produce garbage whatever tool set they have to work with. Don't blame a hammer when you hit your thumb.
-
Unfortunately ML neural nets are descendents of Igor Alexander's WISARD. That distinguished well between cars and tanks in the lab, but failed dismally in the field. Eventually they realised it had trained itself to distinguish between cloudy and sunny days. It is said colleagues then refused to acknowledge Alexander's presence on sunny days :)
Quoting a famous incompetently engineered example doesn't really say anything useful. Those people were arrogant idiots. Imperial seems to produce more than its fair share of those.
There are documented examples where changing one pixel in a photo of a road "stop" sign, caused the ML system to change to interpreting it as a "40MPH" sign.
This is an issue with current cars. Sometimes a very dirty and unclear sign fools them, and you can understand it. Other times you can drive through an area of road works with numerous speed restriction signs, all nice and clean and clear, and the car misreads every one in the same odd way. As of this month cars sold in the UK are required to have a road sign reader nagging the driver. Some wanted that to enforce the speed restriction, which thankfully didn't get through.
ML radiography interpreters was found to be deciding whether or not a cancer was treatable based on the font used in the radiograph. The training set correctly detected that one font was associated with a hospital in a poor area where the treatment outcomes were poorer.
ML in the (US) judicial system decides that black people shouldn't be released on bail, since that is the prejudice embodied in the training set.
Then there are the more subtle legal consequences - if a company using ML for employment decisions is accused of racial bias, they won't be able to disprove it.
FFI and references, see comp.risks; the 40 year archive is at http://catless.ncl.ac.uk/Risks/ (http://catless.ncl.ac.uk/Risks/)
There are a lot of poor quality engineers out there. They'll produce garbage whatever tool set they have to work with. Don't blame a hammer when you hit your thumb.
These problems are inherent in ML systems.
With neural nets all you have is vast numbers of randomly interconnected multiplier accumulators and registers. You throw a training set at them, and they somehow assign interconnections and weighting factors. There is no reason to believe that if you present the training set in a different order, you will get the same interconnections and weighting factors.
With a conventionally engineered system you can insert diagnostic mechanisms to indicate why the system is doing something. With ML systems you can dump the weighting factors and interconnections, and nothing else. That is equivalent to trying to understand human thinking by examining neurons firing.
The major problem with neural nets is that the ignorant are looking for quick fixes, and they believe the salesmen/advocates. Given that the ignorant don't understand engineering, it isn't surprising they can't tell the difference between "designed magic" and "found magic".
Even if you find an ML system produces the "right" output with the current set of tests you've thrown at it, there can be no way of telling how it will react to the next test. That's the antithesis of engineering.
I have no problems with ML being used to generate pictures of WarCraft characters or Doctor Who trailers: the result goes through a persons mind before acceptance/rejection. Not so with hire/fire incarcerate/release treat/leave stop/go decisions :(
-
These problems are inherent in ML systems.
These are problems with all learning. There can be a fine line between education and indoctrination.
With neural nets all you have is vast numbers of randomly interconnected multiplier accumulators and registers. You throw a training set at them, and they somehow assign interconnections and weighting factors. There is no reason to believe that if you present the training set in a different order, you will get the same interconnections and weighting factors.
Just like teaching a human.
With a conventionally engineered system you can insert diagnostic mechanisms to indicate why the system is doing something. With ML systems you can dump the weighting factors and interconnections, and nothing else. That is equivalent to trying to understand human thinking by examining neurons firing.
A fully deterministic system is more predictable, but we currently have to use humans for many tasks, because we need a level of flexibility those deterministic systems can't offer.
The major problem with neural nets is that the ignorant are looking for quick fixes, and they believe the salesmen/advocates. Given that the ignorant don't understand engineering, it isn't surprising they can't tell the difference between "designed magic" and "found magic".
Just like every new thing that comes along. Humans are serial technology abusers.
Even if you find an ML system produces the "right" output with the current set of tests you've thrown at it, there can be no way of telling how it will react to the next test. That's the antithesis of engineering.
If you teach a human to do something, you have little idea how well it went, and whether they will handle anything outside their learning in an acceptable manner. At least the ML system forgets a lot less.
I have no problems with ML being used to generate pictures of WarCraft characters or Doctor Who trailers: the result goes through a persons mind before acceptance/rejection.
Recently we seem to have seen numerous fiascos where there was no human WTF moment resulting in a cleanup before release.
Not so with hire/fire incarcerate/release treat/leave stop/go decisions :(
Hire and fire is worse than an ML problem. although there is one. The human element is HR staff, whose ability in talent spotting usually seem to be to reject it. All the best candidate can be found in their waste bin.
-
These problems are inherent in ML systems.
These are problems with all learning. There can be a fine line between education and indoctrination.
With neural nets all you have is vast numbers of randomly interconnected multiplier accumulators and registers. You throw a training set at them, and they somehow assign interconnections and weighting factors. There is no reason to believe that if you present the training set in a different order, you will get the same interconnections and weighting factors.
Just like teaching a human.
With a conventionally engineered system you can insert diagnostic mechanisms to indicate why the system is doing something. With ML systems you can dump the weighting factors and interconnections, and nothing else. That is equivalent to trying to understand human thinking by examining neurons firing.
A fully deterministic system is more predictable, but we currently have to use humans for many tasks, because we need a level of flexibility those deterministic systems can't offer.
The major problem with neural nets is that the ignorant are looking for quick fixes, and they believe the salesmen/advocates. Given that the ignorant don't understand engineering, it isn't surprising they can't tell the difference between "designed magic" and "found magic".
Just like every new thing that comes along. Humans are serial technology abusers.
Even if you find an ML system produces the "right" output with the current set of tests you've thrown at it, there can be no way of telling how it will react to the next test. That's the antithesis of engineering.
If you teach a human to do something, you have little idea how well it went, and whether they will handle anything outside their learning in an acceptable manner. At least the ML system forgets a lot less.
You can ask a human subtle oblique questions to see how they reached a decision. You can't do that with an ML system.
ML systems do "forget". All it needs is:
- you spot a problem in an ML's output
- you apply more training examples in the hope they will reconfigure some of the pathways and weights
- you cannot have any concept of how the new pathways/weights will change previously correct output. That's a real problem
I have no problems with ML being used to generate pictures of WarCraft characters or Doctor Who trailers: the result goes through a persons mind before acceptance/rejection.
Recently we seem to have seen numerous fiascos where there was no human WTF moment resulting in a cleanup before release.
Not so with hire/fire incarcerate/release treat/leave stop/go decisions :(
Hire and fire is worse than an ML problem. although there is one. The human element is HR staff, whose ability in talent spotting usually seem to be to reject it. All the best candidate can be found in their waste bin.
HR droids want to offload the responsibility for their crap decisions onto a computer. The aren't alone in wanting to armour-plate their backs. So do some magistrates, judges, insurance companies, airlines (see the recent Air Canada debacle!).
-
You can ask a human subtle oblique questions to see how they reached a decision. You can't do that with an ML system.
You can query a well designed ML system, too. You can probe what lead to its conclusions, and probably get more meaningful output than from the average pleb.
ML systems do "forget". All it needs is:
- you spot a problem in an ML's output
- you apply more training examples in the hope they will reconfigure some of the pathways and weights
- you cannot have any concept of how the new pathways/weights will change previously correct output. That's a real problem
We don't usually classify an update in thinking based on new information as forgetting. You can't control a human's learning updates, but its easy to lock down an ML solution if you want to.
-
You can ask a human subtle oblique questions to see how they reached a decision. You can't do that with an ML system.
You can query a well designed ML system, too. You can probe what lead to its conclusions, and probably get more meaningful output than from the average pleb.
Reference please. Or are you relying on "well designed" as being the weasel-words.
With people you can probe the mental model they have of the problem. I accept sometimes it will be just "mental", but that is in itself an adequate result! :)
You cannot even manage that with ML systems since they do not have an identifiable mental model per se; they just have neurons and weighting factors.
ML systems do "forget". All it needs is:
- you spot a problem in an ML's output
- you apply more training examples in the hope they will reconfigure some of the pathways and weights
- you cannot have any concept of how the new pathways/weights will change previously correct output. That's a real problem
We don't usually classify an update in thinking based on new information as forgetting. You can't control a human's learning updates, but its easy to lock down an ML solution if you want to.
In practice you can't lock down an ML system, because there will always be the requirement to remove newly-discovered edge cases. And there will always be newly-discovered edge cases.
Over-the-air updates are already a problem with driverless cars, because when you get in a car you can't rely on it behaving the same way that it did yesterday.
Have you - like most young software "engineers" - forgotten that "You can't test quality into a product". When you put that to people creating ML systems based on training sets (i.e. all of them), first they pull a face, then they go "la-la-la-la-la".
-
You can ask a human subtle oblique questions to see how they reached a decision. You can't do that with an ML system.
You can query a well designed ML system, too. You can probe what lead to its conclusions, and probably get more meaningful output than from the average pleb.
Reference please. Or are you relying on "well designed" as being the weasel-words.
Most ML systems are split into a training system, and a much simpler run time system. The run times are as basic as possible, but the systems on which the learning takes place generally have pretty flexible facilities for getting an explanation for a decision.
With people you can probe the mental model they have of the problem. I accept sometimes it will be just "mental", but that is in itself an adequate result! :)
You cannot even manage that with ML systems since they do not have an identifiable mental model per se; they just have neurons and weighting factors.
ML systems do "forget". All it needs is:
- you spot a problem in an ML's output
- you apply more training examples in the hope they will reconfigure some of the pathways and weights
- you cannot have any concept of how the new pathways/weights will change previously correct output. That's a real problem
We don't usually classify an update in thinking based on new information as forgetting. You can't control a human's learning updates, but its easy to lock down an ML solution if you want to.
In practice you can't lock down an ML system, because there will always be the requirement to remove newly-discovered edge cases. And there will always be newly-discovered edge cases.
Most ML systems can't be trained, as they only perform the run time aspects of the problem. So, any updates come from the learning system, and the transfer of updates from there to the numerous run time systems can be as orderly or chaotic as you make it.
Over-the-air updates are already a problem with driverless cars, because when you get in a car you can't rely on it behaving the same way that it did yesterday.
That is a management problem. Its a pretty dumb process that pushes updates randomly, and surprises a driver in the middle of a journey with new behaviour, whether its an ML behaviour, or something altered in the car's UI.
Have you - like most young software "engineers" - forgotten that "You can't test quality into a product". When you put that to people creating ML systems based on training sets (i.e. all of them), first they pull a face, then they go "la-la-la-la-la".
You can't test quality into a product, but also you can't built a complex product to be perfect. That always defeats human capabilities. Even the best engineered systems do things that surprise their designers, and take years to fully shake out. There is plenty of denial about the vast amount of work it will take to get, say, a driverless car that needs one or two human interventions per journey to one that might be no more dangerous left on its own than the average human. People find it hard to face the reality of just how difficult awkward cases are compared to more straightforward ones. The sort who can't accept just how hard it would be to properly automate the office cleaner's job, and don't understand that the office cleaner is most likely to be laid off because everyone else in the office has been eliminated from their jobs.
-
You can ask a human subtle oblique questions to see how they reached a decision. You can't do that with an ML system.
You can query a well designed ML system, too. You can probe what lead to its conclusions, and probably get more meaningful output than from the average pleb.
Reference please. Or are you relying on "well designed" as being the weasel-words.
Most ML systems are split into a training system, and a much simpler run time system. The run times are as basic as possible, but the systems on which the learning takes place generally have pretty flexible facilities for getting an explanation for a decision.
Please provide references about those "facilities for getting an explanation". Fundamentally it is a serious research topic, which no clear resolutions on the horizon. It will be a fruitful source of PhDs and research grants for decades.
Even where the rules were explicitly coded (i.e. 1980s old-skool AI), in practice it was difficult to determine why a decision was made, and then to modify the rules to make the desired change and no other. In modern ML systems there are no explicitly coded rules, so even that doesn't work.
I'm sure ML systems they are structured that way: the training systems determine the weights and interconnections, and the deployed runtime executes them. Hence you can see that distinction is irrelevant to the points I've been making.
The second reason it is irrelevant is that deployed systems don't necessarily retain all the information that caused them to make a decision, so their decision making process can't be "replayed" back in the lab.
With people you can probe the mental model they have of the problem. I accept sometimes it will be just "mental", but that is in itself an adequate result! :)
You cannot even manage that with ML systems since they do not have an identifiable mental model per se; they just have neurons and weighting factors.
ML systems do "forget". All it needs is:
- you spot a problem in an ML's output
- you apply more training examples in the hope they will reconfigure some of the pathways and weights
- you cannot have any concept of how the new pathways/weights will change previously correct output. That's a real problem
We don't usually classify an update in thinking based on new information as forgetting. You can't control a human's learning updates, but its easy to lock down an ML solution if you want to.
In practice you can't lock down an ML system, because there will always be the requirement to remove newly-discovered edge cases. And there will always be newly-discovered edge cases.
Most ML systems can't be trained, as they only perform the run time aspects of the problem. So, any updates come from the learning system, and the transfer of updates from there to the numerous run time systems can be as orderly or chaotic as you make it.
ML, in the modern meaning of the words, are always "trained by rote" on many many individual examples. Old-skool AI systems were "taught by general rules".
Over-the-air updates are already a problem with driverless cars, because when you get in a car you can't rely on it behaving the same way that it did yesterday.
That is a management problem. Its a pretty dumb process that pushes updates randomly, and surprises a driver in the middle of a journey with new behaviour, whether its an ML behaviour, or something altered in the car's UI.
No, it is a technical problem and a user problem.
Management ought to remove/avoid such problems, but in practice they are only to eager to turn a blind eye.
Have you - like most young software "engineers" - forgotten that "You can't test quality into a product". When you put that to people creating ML systems based on training sets (i.e. all of them), first they pull a face, then they go "la-la-la-la-la".
You can't test quality into a product, but also you can't built a complex product to be perfect. That always defeats human capabilities. Even the best engineered systems do things that surprise their designers, and take years to fully shake out. There is plenty of denial about the vast amount of work it will take to get, say, a driverless car that needs one or two human interventions per journey to one that might be no more dangerous left on its own than the average human. People find it hard to face the reality of just how difficult awkward cases are compared to more straightforward ones. The sort who can't accept just how hard it would be to properly automate the office cleaner's job, and don't understand that the office cleaner is most likely to be laid off because everyone else in the office has been eliminated from their jobs.
Quite right: the first 50% is easy, the last 20% extremely difficult. 50% is acceptable for WarCraft character generation, but is completely inadequate for important irreversible decisions (e.g. judicial imprisonment, medical diagnosis/treatment, autonomous vehicles, etc).
You are making my points for me. Thanks.
Do us all a favour, and subscribe to comp.risks. It is low-volume and high-quality curated information source - with a 40 year pedigree to prove it!
EDIT: the RSS feed is http://catless.ncl.ac.uk/risksrss2.xml (http://catless.ncl.ac.uk/risksrss2.xml) but there is also a usenet feed and I expect you can get an email (about 2 per week).
-
Though the few PhD defenses I've recently attended are all of the form "we used a neural network and magically solved all our problems"... ::)
... to which the next questions are "how can you prove the problem was completely solved?" followed by "where is the boundary where the problem was no longer completely solved?".
ML inherently has problems explaining why, and there are numerous examples where a tiny change caused failure.
If I were a young Turk looking for a research field today, I'd probably look into ways to build robust systems from powerful but imperfect components whose flaws are not always discoverable a priori, and not always correctable when they are.
Because that's how everything is going to work from now on, pretty much.
-
Though the few PhD defenses I've recently attended are all of the form "we used a neural network and magically solved all our problems"... ::)
... to which the next questions are "how can you prove the problem was completely solved?" followed by "where is the boundary where the problem was no longer completely solved?".
ML inherently has problems explaining why, and there are numerous examples where a tiny change caused failure.
If I were a young Turk looking for a research field today, I'd probably look into ways to build robust systems from powerful but imperfect components whose flaws are not always discoverable a priori, and not always correctable when they are.
Because that's how everything is going to work from now on, pretty much.
It is an excellent field for fundamental research. Unfortunately systems are deployed without that :(
Fundamentally there have been no theoretical breakthroughs in the past 40 years (arguably 60).
For ML systems we haven't even reached the "peak of inflated expectations" yet, let alone the "trough of disillusionment". Hence it is impossible to know when - or even if - we might get through the "slope of enlightenment" to the "plateau of productivity".
-
The real way to find out what you do for a PhD nowdays is to read dissertations. Several schools publish the dissertations of their students, and others aren't that hard to obtain. In general they are not new theory, but applications of theory to specific applications. An example that is now a few decades old is application of control theory (observability and controllability specifically) to economic systems. I would agree that there hasn't been much if any groundbreaking new theory on the level of Maxwell, Shannon or the like but that doesn't indicate that everything is known. Lots of room left in non-linear systems, time varying systems and the like. Engineering of the newer high temp superconductor materials into real applications is another obvious area. Another example of application that has taken off is the thermal bolometer focal plane array. No new theory, but forty years ago you could have found numerous experts in the infrared field that would have denied the possibility of such a critter. Taking that from idea to reality generated and/or utilized quite a few PhDs.
-
Another example of application that has taken off is the thermal bolometer focal plane array. No new theory, but forty years ago you could have found numerous experts in the infrared field that would have denied the possibility of such a critter. Taking that from idea to reality generated and/or utilized quite a few PhDs.
Recall seeing Dr Wood's early uncooled bolometer arrays at work at Honeywell Corporate Technology Center in the 1980s, sensitive enough to not only spot a tank hiding in the trees & bushes, but see the thermal tracks it left for the pervious 1/2 hour getting there!!
Later we utilized cooled HgCdTe 8-12 micron detectors to map the chemical atmospheric characteristics passively 5 KM away for Remote Chemical Agent Detection (XM21). These detectors were sensitive enough to detect the heat of a human body at 1000 miles!!
Lots of early work in these fields and plenty of opportunities for young PhD students!!
Today, the most opportunity likely exist with the semiconductor field, as they seem to be constantly changing physics on a yearly basis with the continual node reductions, as "You Can't Do That!!" seems to be their Marching Song :-+
Still can't gather how quickly these latest semiconductor advancements ends up in the general public hands ;D
Hats off to them, and hope they continue :clap:
Best
-
Though the few PhD defenses I've recently attended are all of the form "we used a neural network and magically solved all our problems"... ::)
I am not in any academic field at all, but this sentence makes me curious: Are defenses like that actually successfull?
The closest I've seen is when the student nominally passes, but needs to show some additional work prior to getting their degree. As a 'general public' attendee of the defense, I don't get to verify that, but the student's PhD committee does. At least that's how it works at my university.
Though the few PhD defenses I've recently attended are all of the form "we used a neural network and magically solved all our problems"... ::)
... to which the next questions are "how can you prove the problem was completely solved?" followed by "where is the boundary where the problem was no longer completely solved?".
ML inherently has problems explaining why, and there are numerous examples where a tiny change caused failure.
If I were a young Turk looking for a research field today, I'd probably look into ways to build robust systems from powerful but imperfect components whose flaws are not always discoverable a priori, and not always correctable when they are.
Because that's how everything is going to work from now on, pretty much.
That approach has merit in the general case, though in my specific field (remote sensing) it's often hard to get 'ground truth' data. You often see years of PhDs design ML systems (today) and fuzzy logic classifiers (back when I was in the thick of it) using a handful of verification datasets since they are so hard to come by (for example: fly a sampler probe into a thunderstorm using an armor-plated aircraft (https://explore.digitalsd.org/digital/collection/T28)). In other cases, the verification data is easier to measure, but harder to qualify (as in, is the remote sensing data and the ground truth data coming from the same physical phenomenon, either temporally or spatially?) A good fraction of most such theses deal with these issues, though there's often a fair bit of hand-waving.