Contemplating about a second PhD in statistics/machine learning

Question

I have been thinking about a second PhD for the last year. I am currently a doctoral candidate in civil engineering at University of Texas at Austin (ranked 6th in my discipline) and working as a student statistical consultant at the university consulting center. I am working on spatio-temporal modeling of count data using Bayesian hierarchical models, with computationally efficient techniques for my dissertation. (I will have publications on this very soon; two of them are in review). I do have 2 publications in my area but their topic is the application of statistical models. I am also going to get a Masters in statistics next semester along with my PhD in civil engineering. I have done many courses related to Bayesian statistics including graduate level mathematical statistics, theoretical MCMC, stochastic volatility and time series models, statistical consulting, advanced econometrics (non-Bayesian), discrete choice modeling and one course on data mining (graduate level). In my field, I see massive datasets but very minimal statistical expertise, particularly on the big data side. This has motivated me to pursue something beyond my Phd and beyond my discipline.

I am very interested in handling large datasets and perhaps, machine learning applications. I would like to know whether a PhD in machine learning is going to help me realise my dreams. I do not have any formal research experience in data mining or machine learning. But, I do a lot of Bayesian hierarchical modeling on smaller datasets. Given my experience, I am not sure whether I can secure admission to a good program in the machine learning area. I appreciate any suggestions and advice on whether to pursue another PhD and the feasibility of securing admission to a good program in a PhD machine learning track. I am assuming basic financial assistance for any PhD program.

Thanks much in advance.

Answer 1

As other people already stated, multiple PhD's is almost certainly not a good idea. How to transition into machine learning depends on where you want to work:

• academia, you can try to get a postdoc position in machine learning, in that way you can build your career by writing publications and get up to speed with machine learning during your postdoc research. This could be problematic as you might have to compete with people that have a PhD in machine learning. But this can be overcome during the interview stage.
• industry, you can simply apply to entry level machine learning positions (look for data science positions) and take it from there. Once you convince a company that you are a worthwhile addition, you can learn the ropes on the job.

In general, I would recommend already getting up-to-speed a bit in machine learning, try some online tutorials, etc. This can really help in securing a new position in either academia or industry.

Answer 2

This site has had such questions come up before. Usually (as in this case), I think the correct answer is that doing a second PhD is not necessary, and not a good use of your time. In my opinion, shared by many (most?), multiple PhDs are very rarely a good idea. As people have remarked elsewhere, a big part of a PhD is learning how to do research (hopefully) under supervision and guidance. The supervision and guidance do not necessarily happen in practice, anyway. Usually much of a PhD is deadwood and bureaucracy, like required courses.

My personal opinion are that PhDs are not a necessary qualification in any case. Before they existed people did just fine. Check out the history of the PhD on Wikipedia

If you know enough statistics to write research papers, you just go ahead and write papers. Once you know enough to do so, it is not rocket science. Machine learning is just statistics done by computer scientists for some reason, maybe because the statisticians are not interested in doing it.

I think your general aims and perspective are sensible. Knowing things like Bayesian statistics and modelling is useful when working with data.

In my field, I see massive datasets but very minimal statistical expertise, particularly on the big data side.

Yes, this sounds accurate. Statistical expertise is not very commonly available or applied by non-statisticians, but it would be useful to them if they knew how to do it correctly.

A couple of suggestions wrt things you could usefully focus on:

1. English language skills. Very important for a researcher. This is particularly important if one does not speak English natively.

2. Computational skills. This is getting to be also a truism, but people in academia don't usually know what they are doing re programming and software development. A better understanding in these areas will probably pay dividends for researchers in the applied sciences, though there are differences of opinion as to how much.

Neither of these skill sets would be usefully served by getting a second PhD.

I have a PhD in Statistics, but I don't think I need to have one to offer the preceding opinions. Mostly they are just common sense.