An Interview with a Springboard Machine Learning Expert Daniel CarrollBy: Roger Huang, Springboard
Daniel Carroll is a principal data scientist at Aetna, one of largest health insurance companies in the world, with about 50,000 employees. As he writes in his LinkedIn bio, "Artificial intelligence is everywhere today, my area of focus is application of neural networks to unstructured data: text, imagery, sensor output. These days I'm working on a variety of healthcare problems using AI. In the past I've worked in the defense industry developing AI and analytics based solutions to tackle some of our nation's toughest problems."
Daniel has started mentoring students in the AI/Machine Learning Career Track Springboard offers, the first to offer a personalized education in machine learning with weekly expert video calls and a job guarantee. We sat down to talk to him about his experience with the program so far, and insights he has on a machine learning career path.
1) Tell me your story. How did you get into machine learning?
I started in college, through the selection of my major (mathematics). I didn't have real exposure to it until I worked at a government lab in school. While I was there, I was exposed to machine learning and neural nets and it blew my mind—I knew right away I needed to get involved.
I then worked on a DARPA project and then worked on various government projects, mostly machine learning related, in the next few years. I would do things like figuring out optimizations for aircraft allocation over battle space.
Two years ago, I moved to Aetna, a health insurance company. It was a breath of fresh air. I'm a principal now, which means I get lots of independence on projects such as the application of natural language processing to search results.
To summarize, I discovered machine learning pretty early on through school and have been obsessed with it ever since.
2) What's your day-to-day like at Aetna?
I'm primarily focused on ideation and getting buy-in from our decision-makers. I do everything from strategy to individual contributor work. Let me give you an example.
A real problem in healthcare is that people are constantly looking for high-quality care. A lot of this is a translation problem: the medical industry has a way of describing things that is hard to follow for most people. It is a difficult system to navigate. Promoting literacy on the healthcare system and implementing a more natural search experience can help people find the care they're looking for.
I put together a model to map user queries to relevant care using a combination of deep learning and some more traditional methods. This model allowed us to achieve far better results than the standard keyword or text search technique we were using at the time. Before I bothered trying to get it into production, I built it into a web application and demoed it internally to the leadership team. The best way to get buy-in is to show people your model in action even if it's rough.
A lot of my day-to-day is spent talking with other software engineering teams about how to make machine learning models integrated with their work, as well as trackable and performant. Most of the effort is not on models, it's on ancillary tasks. One of the biggest tasks (which is covered in Springboard's AI/Machine Learning Career Track) involves putting a model into production and making sure it is testable and integrated to the point where it performs like general software.
Finally, I do participate a lot in the hiring process, both finding and selecting candidates, and defining what the hiring pipeline looks like. Finding good people is a huge part of the job -- it's actually my favorite part!
3) Why did you want to mentor for Springboard's Machine Learning Career Track?
I got contacted about it by a recruiter. I was really taken aback by how good the curriculum is -- it seemed reverse-engineered to the needs I have as a hiring manager. I had just come from talking to applied math PhDs about getting a job in industry. Most of the things I told the students to do was built straight into the curriculum.
4) How has your experience been so far with mentoring?
I have loved it so far -- the quality and attitude of the candidates in the Springboard program is the most astounding thing. Take the top percentile of any academic class I've been in and collect them together - that's what the Springboard AI/Machine Learning Career Track feels like.
One of my students has a PhD in applied math and has been doing quantitative trading for a decade. I feel like I have as many questions for him about that as he does for me about ML. The conversations I have with them are very illuminating. The students are addictive -- their energy has inspired me to learn more.
I've also done some TAing for the course, so I've met some of my fellow mentors and other students in the course and I can say this attitude generalizes to them.
5) What pieces of advice would you give people looking to get into machine learning roles?
Let's split this in two.
First, career advice if you're looking to become a machine learning engineer. A lot of people like to focus on the 10% (the machine learning and fitting component) but that's a very small percentage of the work that needs to be done. Working on machine learning research requires you to get a PhD to solve for fundamental ML problems and is a much smaller world than most people realize. If you want to become a machine learning engineer, having a Swiss Army knife skillset works better. If you understand enough about how to build a web app, how to write performant software, how to interact and explain abstract ideas, and you are capable of working across a broad spectrum of problems, hiring managers want that. ML is shifting from abstract ideas to engineering. Building projects will help you stand out.
Second, if you're just curious about learning machine learning: don't neglect the fundamentals. It is easier than ever to implement neural nets thanks to the cheapness of modern computing and easy to use frameworks, which is a double-edged sword. You can skip a lot of things and go straight to results without really understanding what you're doing. I recommended the book Deep Learning by Ian Goodfellow to really dive deep into the basics and make sure you have a solid foundational understanding.
6) Is there a bias against hiring bootcamp grads for machine learning roles?
So I have just a bachelor's degree, but that's definitely not the norm. Many of the best data scientists and AI people I have worked with have been PhDs. There certainly is a bias towards having an advanced quantitative degree, and I think it's well deserved. There can be a bias against people with only bootcamp experience as well. However, the one thing a bootcamp can deliver that a really good university can't is a really thorough understanding of how to build something from ideation to deployment -- that's a full-fledged capability I don't see often from PhDs. If you can bring that to light in a CV or interview, that will make you shine.
Use your portfolio projects as a stamp and ask your interviewer to play around with the model you built and deployed online. Take out a laptop during the interview and show a model in production to your interviewer. That's so impressive to me. Demonstrating bootcamp experience through projects helps derisk you from a hiring manager perspective: it shows that you can build something from start to finish.
If you're interested in being mentored by somebody like Daniel and being supported by a job guarantee, look no further than Springboard's AI/Machine Learning Career Track.
This post was sponsored by Springboard. To learn more about Springboard, check out their reviews on SwitchUp.