Help me pick out my safety schools (for machine learning)?

[sisyphus1]

Quick Background

 

GRE (Q/V/W): 166/167/5.5

Undergrad: Physics/Econ double major from a top 10 school (GPA=3.8)

Grad: MS in computer science from Columbia (GPA=4.0)

Publications: 1 first-author at a top conference, 1 first-author at a workshop

Research experience: Nothing other than class projects (the two publications above were as a result of classroom projects). I think this is my biggest weakness

LoR: should be decent, and from people who are well-known. But definitely not "best-student-ever" type LoRs.

 

My research interest is in Machine Learning.

 

I'll be applying to 10~12 schools, including the top 4. I think I have a good list of reach/match schools, but I am having a hard time choosing the "safety" schools. With the caveat that there are no true "safety" schools when it comes to PhD admissions, can people recommend some safety schools that are also good at ML?

 

Thanks!

 

[D3veate]

Here's a strategy that I haven't heard mentioned: your bullet points look excellent, so why don't you ignore safety schools? If you don't get in, you can spend the year in industry and try again in a year. A year in industry won't hurt your application, and if you demonstrate that you're able to pursue independent research, your application may be even stronger.

My rational for suggesting this is:

1) Safety schools are necessary for high-school graduates, but someone with a Master's degree has more options.

2) The prevailing thought is that the probability of being accepted into a top-tier university is roughly equal to the published acceptance rates. Everyone is qualified, so it's appearantly random what makes it so that one person is accepted while another is not. If the acceptance rate is, say, 0.1, and the results one year are independent of results on subsequent years, then the probabiliy of being accepted into one of 10 universities is $1 - (1 - 0.1)^{10} \approx 0.65$. If you apply to the same universities for two years in a row, that becomes $1 - (1 - 0.1)^{2 \times 10} \approx 0.88$. I've heard that spending up to 3 years in industry doesn't hurt your application (aside from making it harder to get good recommendation letters). If you follow this strategy for 3 years, you should be able to apply during 4 application seasons. That becomes $1 - (1 - 0.1)^{4 \times 10} \approx 0.99$.

[Icydubloon]

Safety School = Your Undergrad School

[Kleene]

Safety School = Your Undergrad School

This is what I am planning on doing. Don't you think it will be very awkward to ask for a reference from your previous (undergrad) supervisor for other institutions if you have that same person as a back-up PI? "I really like(d) working with you, but I hope you don't mind you are last on my list."

[sisyphus1]

Safety School = Your Undergrad School

 

Any reason for this? My undergrad is very highly ranked for machine learning and I would imagine it would be a reach for me.

[milara]

Why are you drawing a distinction between research that originated in a class and research that did not? Unless your publications based on the class work were in student-only CfPs, whether it originated in a class or not is irrelevant. You did something that your peers determined was worth publishing.

 

As for safety schools... I suppose I can suggest my school, Northwestern. We have some coursework that might interest you. We have a class in sparse and low-rank recovery problems that I took which was excellent, albeit experimental, since there are no textbooks on the topic. There's a class in probabilistic graphical models which I didn't take, but which I've heard good things about. We're also one of the schools that is getting access to IBM Watson for a projects class... I don't know if that's a one-off thing or if it will be offered again, but it's worth mentioning.

 

When you say your research interest is machine learning, do you mean that you need to learn machine learning, or that you already know machine learning pretty well and want to apply it to research? And, do you want to be developing new methods, or applying it in innovative ways, or both?