Industry vs Academia in Machine Learning | by Purvanshi Mehta – Towards Data Science

Sign up
Sign in
Sign up
Sign in
Purvanshi Mehta
Follow
Towards Data Science
—
4
Listen
Share
Choosing careers in itself is difficult. Add the time commitment of 5–6 years for a PhD as compared to the lucrative industry job and your mind boggles with the possibilities. I was in a similar position after my undergrad and master’s (yes both times).
If you know my career path I was luckily introduced to research during my sophomore summer where I worked on semi supervised relation extraction. Later I worked full time at a research lab before joining grad school. This gave me a glimpse of the life as a full time researcher / PhD student.
But despite the research experience I joined Microsoft (where I am currently working) after my Master’s. The process of decision making was tedious and therefore I finally pen down my thoughts. I highly recommend to take this post as an opinion(and more of how and why I took this decision) and talk to multiple people in industry and academia before finalizing yours.
One thing I have realized after working full time both in academia and industry is the fact that
Industry or academia you get to work on exciting stuff and are a part of the AI revolution. Now it’s just a matter of which level of the advancement you want to work on.
You want to work on porting an age old rule based system to ML/DL models? Or scale the existing ML regression model to something more state of the art? Or you could work on more fundamental problems and try to answer questions which would help in building systems not immediately but in the next 5–10 years(obviously depending on the application). I tried to answer the part of which ‘advancement level’ I want to work on by asking the following questions to myself~
Industrial research is mostly product oriented
unless you are in a pure research group (which is difficult to get into just after a Master’s or even after a PhD). Usually you have a problem statement related to the product and you try to find related work about how to solve it. If you find something novel in the process, you might go ahead and publish it.
Most groups in the industry do not consider published work in their performance review process. So basically you have to be internally motivated yourself.
If you are not fixated on the problem statement then industry research groups might be a good option.
For me I found a group which was actively publishing and also had majority of people with PhDs in it. Therefore I found it a good fit for myself.
ML research is interesting but not all research is impactful. Let’s look at the numbers ~
The number of AI-related publications on arXiv grew by more than sixfold, from 5,478 in 2015 to 34,736 in 2020 . Quantity is definitely on the rise but I am not sure about the quality.
The following is a graph of NeuriPs paper citations per annum. As the number of publications increase, the average number of citations to your publications decrease with every year. That means if you were publishing at Neurips in 2017 your average number of citations or the number of people using your work would be 4.6. These numbers look pretty grim to me.
I am aware there are exceptional papers and average citations is not the greatest measure to measure a paper’s worth but atleast an approximate evaluation of your impact.
There are exceptional contributions (which actually drive the field) but there are a lot of factors which would have contributed to me writing such papers(some of which being advisor, which research lab I am in, research area and great mental health throughout my PhD).
In 2019, 65% of graduating North American PhDs in AI went into industry — up from 44.4% in 2010, highlighting the greater role industry has begun to play in AI development.
When I was working at Amazon as an applied Scientist all the people in my team were PhDs. As I interacted with them more and more I realized I was and could do similar work as a Master’s student.
The boom in the AI development in the past few years has been phenomenal. There’s a lot of engineering to be done to make great products which requires the skill of reading papers and thinking of a use case to apply them or you have a use case and read papers which would solve the problem.
The truth is not everyone is doing just research after getting a PhD in Machine Learning.
What if I would be working in a similar group as I am now after getting my PhD? Going through the whole process didn’t make sense then.
Let’s be honest here- Academia is difficult.
Irrespective I highly appreciate people who work in academia despite of all these factors. I understand the quest for novel ideas and working on fundamental problems. I can empathize the intellectual simulation when you get to the bottom of a problem!
I still think about going back to academia, specially when I have to do mundane tasks in my everyday job (welcome to industry). There are many benefits of the work you do as a PhD
At the end it’s all about working on exciting topics. Be that be in production or in research. I don’t know if I will ever get a PhD. But if I do, I would definitely be better prepared. Industry will add much needed perspective to what matters and what are the important problems.
—
—
4
Towards Data Science
Graph Intelligence @Microsoft
Help
Status
About
Careers
Blog
Privacy
Terms
Text to speech
Teams

source

About The Author

admin

See author's posts