Statistics to CS PhD

[qqyyzz]

I'm currently in a stat PhD program and I just received my MS last spring from a top 20 ranked statistics program.  I know that graduate students switching programs is relatively rare so I was hoping to get some feedback from experts on this forum on feasibility.  If I were to switch over I would mainly interested in a PhD in CS focused on machine learning and data mining.

 

Machine learning is studied in both computer science and statistics.  Statistics takes more of a modelling approach to data while computer science takes a more algorithmic approach (for more details, see Leo Breimans paper: Statistical modelling a tale of two cultures).  My main desire to switch is due to the cultural differences in how CS versus statistics treats machine learning.

 

Many statistics departments have a very conservative culture.  As an example, Andrew Gelman's blog mentions that many statistics departments have a measure theory requirement that is perhaps outdated.  While it's useful for theorists he argues that for applied statisticians in a modern environment computing requirements would be much more useful.  Additionally when it comes to machine learning, the statistics departments will generally be much more theory oriented.

 

 

I'm not sure about how easy it would be to get into a CS program with my background.  I come from mainly a math/econ background with a handful of CS courses.  Here's a quick summary of my math, stat, CS courswork:

 

Undergrad math coursework: calculus sequence, linear algebra, discrete math, ODE, numerical analysis sequence, advanced calculus (analysis), abstract algebra, probability, math stats.  [(3.9 GPA for math courses)]

 

Graduate Statistics coursework: theory sequence (casella and berger level), methods sequence (anova, GLM, boostrap, mcmc etc), design of experiments, stat computing, data mining, machine learning and multivariate statistics. [(3.8 GPA for stat coursework)]

 

Computer Science coursework (mix of undergrad/grad courses): data structures, algorithms, pattern recognition (graduate), software engineering, algorithms (graduate - in progress), Artificial Intelligence (graduate - in progress). [(A's in all courses so far)].

 

In terms of research experience, I did have a research project for undergrad with a professor in the econ department on theoretical game theory (the project was funded by external grant if that matters).  I've had a presentation at a conference and I also did a data mining internship this past summer.  Unfortunately no publications (not unusual for stat as most students don't really start research until 3rd or 4th year).

 

I think I have a fairly strong applied math/statistics background (with decent GRE scores, 90+ percentile in math/verbal) but I'm missing some traditional CS courses in programming languages, computer architecture, networking etc.  From the research that I have done computer science programs seem to vary quite a bit in terms of admission requirements and PhD coursework requirement.  What do you think are my chances of getting into a decent machine learning CS program?

Edited September 3, 2014 by qqyyzz

[sisyphus1]

I don't think you have a chance at the top 10, but you may have a good shot at 10-20, if your advisor is supportive of your decision (and write good recs).

 

Here's another thought--why don't you try to find a research group/advisor in machine learning in CS at your current university? If it's a top 20 program, I am sure that the CS department will have at least a couple of ML faculty, and from what I've seen it's not that rare for Stats PhDs to be co-advised by faculty from other departments (e.g. look at student profiles from UCLA/stanford).

[qqyyzz]

I have considered that. Unfortunately the cs department here recently lost their main ML person, and there really aren't any machine learning people left at the cs department. In fact there is only one faculty that is working on anything AI related.

[Edotdl]

CMU has a joint ML/stats PhD program, might be of interest to you, if you haven't looked into it already