Ask HN: If I had to do a phD in ML(CS) which area do I focus on?
I am trying to understand which are the areas in ML that are still in the nascent stages and require extensive research. NLP has been around for half a decade now and is well understand but a few pretty well. CV is picking up now and with the large number of startups it is bound to make great progress.
If I had to start a phD in the near future - which area should I focus on - such that 5-8 years from now it will be commercialized.
I do understand that there is an inherent risk involved. Understanding those risks I am interested to see what other folks think.
2 comments
I graduated with my PhD (in CS/SE) 18 months ago after working in the industry for a very long time before that.
Doing a PhD in ML/CS is a very good idea if you already have a solid understanding. Based on my experiences and observations the most critical success factor is the research group you join and who your supervisory team are. Joining a sub-optimal group (relative to your area of interest) makes the journey much harder and most likely it will take longer too.
My advice is that you track down (Google Scholar is good for this) papers in ML that are of interest to you. Many will be behind paywalls, but you will get an idea of what they cover from the abstracts. While doing this, record the names of the authors and their affiliations. Fairly quickly you will notice who the thought leaders are in your area of interest. Then check out their biographies on their websites. Based on that you should write a proposal for what you are interested in and email them.
If you get the opportunity to keep working (even if part-time) then many research groups will be even more interested in you joining them. Industry-academia collaborations are relatively rare, but much sought after.
You mention NLP. Whilst there have been some impressive breakthroughs, most NLP based systems are guessing using statistics and not applying any logical inferences. That is an area that I have a strong interest in but so far no takers for post-doc work in that area. If you or anybody reading this is interested, then please contact me on my profile's email address.
Doing a PhD in ML/CS is a very good idea if you already have a solid understanding. Based on my experiences and observations the most critical success factor is the research group you join and who your supervisory team are. Joining a sub-optimal group (relative to your area of interest) makes the journey much harder and most likely it will take longer too.
My advice is that you track down (Google Scholar is good for this) papers in ML that are of interest to you. Many will be behind paywalls, but you will get an idea of what they cover from the abstracts. While doing this, record the names of the authors and their affiliations. Fairly quickly you will notice who the thought leaders are in your area of interest. Then check out their biographies on their websites. Based on that you should write a proposal for what you are interested in and email them.
If you get the opportunity to keep working (even if part-time) then many research groups will be even more interested in you joining them. Industry-academia collaborations are relatively rare, but much sought after.
You mention NLP. Whilst there have been some impressive breakthroughs, most NLP based systems are guessing using statistics and not applying any logical inferences. That is an area that I have a strong interest in but so far no takers for post-doc work in that area. If you or anybody reading this is interested, then please contact me on my profile's email address.
Thank you for the reply!
I could not find your email address on your profile. If you could please share - then I had a few more follow up questions and I am curious about the opportunity you have written about.
Thanks again