“Do I need a graduate degree to start a career in machine learning?”
I've been asked this question frequently, and I'd love to answer with an unqualified "No, to get a job in machine learning, you need to know machine learning." But realistically speaking, while the job titles Data Scientist and AI Researcher hold a lot of prestige at the moment, software engineers are in much higher demand, making the transition from software to machine learning more difficult than one might expect. One of the earliest steps for any tech company is "write software." Most of these same companies will not have use for data scientists until they're a bit larger, and similarly, AI researchers are more of a luxury than a necessity for most companies.
Transitioning into a machine learning role can be hard even if you have the right skills (and getting people to agree on what ‘the right skills’ are is its own issue). So it can be helpful to have something that differentiates you from the hoards of other software engineers who want to be doing machine learning.
A master’s degree is definitely one way of doing that. If you are considering a short program (i.e. 1-2 years) at a reputable school, a degree could really help you in a transition. If you manage to do some substantive research while in the program, that could also make your application stand out for a research position as well. However, there are other positions and reources that can be valuable steps toward a career in machine learning.
Data science bootcamps usually provide a review of the machine learning concepts that are most likely to come up in interviews. If you have done work with machine learning before, it might be mostly review. If it is new to you, these 7-8 week programs can move quite quickly. These programs are free to the aspiring data scientist. They make money by charging the hiring company a percentage of the candidate’s starting salary.
Graduate degree required
For those of you who have a MS or PhD that is in something other than statistics, computer science, or machine learning, several of my friends have gone through bootcamps like Insight, have enjoyed their experience, and found the career services helpful for job placement at the end.
No degree required
For those of you who have not spent years getting extensive numbers of degrees, there are still several potential data science training programs available. There are a few programs that focus on creating space or going through a curriculum that teaches Deep Learning. If your goal is ending up in a role that uses deep learning this can be one way to do that.
Insight Artificial Intelligence - This is similar to the other insight programs, but it has a docus on deep learning and reinforcement learning.
fast.ai’s Deep Learning Certificate - In addition to their online course, fast.ai also has an in-person certificate based course available at UCSF with both domestic diversity scholarships and international diversity scholarships.
Insight Data Engineering - If you are open to machne learning adjacent roles, data engineers build the data pipeline before data reaches the data scientists. For this role, you need strong software skills and experience working with data. However, often when many smaller companies say they want a data scientist, what they actually want is someone to set up their data pipeline, so this can still be a great step forward in a transition.
Maybe you do not have the time/resources to move somewhere for a bootcamp, but you already know some programming, have done a bit with machine learning, and want to take a deep dive into a few specific areas or build out a specific project, these are some resources to help make that happen.
OpenAI Scholars - This is a fully funded and mostly completely self-directed fellowship to learn deep learning with at least one hour per week of mentorship from an AI researcher.
AI Grant - Sometimes there is a project that you really want to work on, but you don’t quite have the resources to invest time in it. This grant is for that scenario. It is not a training grant. But that said, there are a lot of really cool machine learning and AI projects that do not require a ton of training to complete. So if you have a specific project you want to work on and you know a little machine learning, you should still apply.
The concept of the AI residency is relatively recent. I saw the first call for applicants to the Google Brain Residency (now renamed the Google AI Residency) around the end of 2014/early 2015, and most of these residencies were announced the following year. This means that we do not yet have a ton of data on where AI residents end up. However, all of the Google AI residents I know have gone on to amazing things, such as staying at Google Brain as a full-time AI researcher or going on to amazing PhD programs. I expect that the current and future AI residents will also go on to very successful careers in machine learning.
Because these positions tend to be pretty selective, having some research experience is usually a major prerequisite. While you might be able to get research during a bachelor’s or master’s degree, if you haven’t managed to do that, don’t despair. Some other options include (a) volunteering to work on a research project at a local university, or (b) trying to replicate the results of a published paper to show that you can follow the steps that someone else took to produce a good research paper.
OpenAI Machine Learning Fellow
I hope this post has given you a better sense of some of the paths into a machine learning career. I will keep it updated as I find more resources. Good luck selecting your next steps!
Written on April 10th, 2018 by Deborah Hanus