The best thing about machine learning jobs?
A lot of them are fully remote. Our guide walks you through the steps you can take to quit cubicle life for good.
This is an installment of our ongoing tech job interview questions series. Today, we’re talking about machine learning interview questions (with a few data science interview questions thrown in).
If you’re reading this, you’ve likely determined that picking up some tech skills is a savvy career move. But once you’ve learned them and updated your resume, how can you feel confident you’ll ace the interview?
This week’s topic goes out to all the Python lovers out there. (How could we not when we teach an in-depth Python course?)
We’ve interviewed a slew of hiring managers, recruiters, and data science team leaders to find out the questions they love to ask machine learning and data science candidates. Use this list to practice the most likely questions (with some curveballs thrown in as well) and you’ll walk into the interview with the confidence of a seasoned pro. Skip to your favorites or scroll the whole way through for more details on how best to answer these.
Table of Contents: 15 Machine Learning Interview Questions
- How would you go about understanding the sorts of mistakes an algorithm makes?
- What data scientists or startups fueled by data science do you admire most and why?
- Give me an example of how you’ve used your data analysis to change behavior. What was the impact, and what would you do differently in retrospect?
- Why are you using that algorithm?
- Why should the business care?
- How do you communicate with both technical and non-technical audiences?
- Explain the steps in making a decision tree.
- What is the goal of A/B testing?
- Why does data cleaning play a vital role in analysis?
- Give me some context around what this set of data was created in, as part of the analytical process.
- Is machine learning a science or an art?
- When is it better to use [insert machine learning library here]?
- How would you remove rain from a short video?
- Give an example of a problem you solved (or tried to solve) with machine learning.
- General technical questions
- Bonus Curveball: The Tank Question
The Best Machine Learning Job Interview Questions
But first, a caveat:
Interviewing strategies (and resulting interview questions) change depending on whether a recruiter is hiring for data scientists, machine learning engineers, or another related position. There’s a good reason for this.
“The interesting thing about… data science versus machine learning [is that] you’re going to get two personas. Data science is quickly being moved into the C-suite while machine learning still relies heavily on coding examples and skill testing.”
This list includes common interview questions for machine learning positions as well as data science positions, so take what you need and leave what you don’t. It will still give you an idea of the types of machine learning job interview questions you can expect—and how you can come fully prepared.
1. How would you go about understanding the sorts of mistakes an algorithm makes?
This question is about showcasing your problem-solving skills beyond catching a mistake in the first place, so focus on actionable steps. It comes from Jason Davis, CEO and Co-Founder of Simon Data, a customer data platform (CDP) provider. Here’s why Davis likes it:
“I want to see that they’re thinking about the problem from multiple angles. At Simon Data, a key part of our offering is the partnership with our client brands to help them solve complex business challenges. We need a candidate who will consider whether the data is bad, or if the algorithm has any unexpected biases that we can address, but the right candidate will also think to look beyond the technical challenge and ask whether we’re even modeling the right business problem for a particular client.”
Bottom line: For questions like these, focus on action, e.g., what next steps you’d take to fix problems, and you’ll stay on track.
2. What data scientists or startups fueled by data science do you admire most and why?
Even if you’re starting in machine learning, you should be aware of the trends and big names in the industry. This question gauges how immersed you are in [data science, machine learning, whatever your focus might be], so the best way to prepare? Have some names (sort of like your favorite influencers) ready to go.
The interview question comes from Jagoda Wieczorek, HR Manager at ResumeLab, who uses it when hiring data scientists—but she often poses a similar question to machine learning engineers as well.
“What’s great about this question is that it allows [us] to put the microscope [to] the candidate’s motivation to become a data scientist and to better understand whether or not they are truly passionate about the field,” she says. “If they can name 2-3 data scientists/startups fueled by data science off the top of their head, that’s when you know you’ve struck real gold.”
Bottom line: Do your research, learn the big names, pick some favorites. FYI, there’s no wrong answer as long as you’re staying up-to-date on industry trends!
3. Give me an example of how you’ve used your data analysis to change behavior. What was the impact, and what would you do differently in retrospect?
This question comes from Melanie Tantingco, RVP of People Ops at Sisense, an analytics platform for businesses and data scientists. “Most people use data to assess decisions—I ask for data to make better ones and change an approach moving forward,” says Tantingco. “I also like to see if our interviewees are reflective and want to be better and improve.”
Bottom line: There’s always room for improvement. You’re not perfect (neither is your work!), so don’t be afraid to focus on how you could do even better next time.
4. Why are you using that algorithm?
As you walk through your process, Susan Shu Chang, a data scientist at Bell says you should expect “Why” questions like this.
“I [usually ask these] in response to what the candidate may have mentioned as an answer in the data science case study. They need to justify it [and show] they know the pros and cons and aren’t just suggesting it because it sounds cool.”
Bottom line: There are lots of choices when it comes to how you approach a problem—you need to show you can think critically about the one you choose to use.
5. Why should the business care?
Another one from Chang (which she admits is a curveball) focuses on whether you can show you understand the bigger picture.
“Optimizing a website with machine learning might be great, but a good answer isn’t ‘because everyone else is doing it,'” says Chang. “It could be something like ‘to save costs from support calls due to the website being better able to address customers’ problems’—we want to see that people know what they are using machine learning for, and understanding the deeper purpose can allow them to solve machine learning problems better.”
Bottom line: Never lose sight of the big picture.
6. How do you communicate with both technical and non-technical audiences?
Sergio Morales Esquivel, a data analytics specialist for Growth Acceleration Partners (GAP) asks questions around communication style every time he interviews for a machine learning role. Why?
“Not only am I looking for applicants who want to share their knowledge and engage with others within our analytics practice, but also those that value the role of transparency and direct communication in problem-solving, discussing alternative solutions and presenting results and insights to peers and other stakeholders,” says Esquivel. “Actively sharing our process… can lead us into avenues of thought that we wouldn’t have arrived at on our own.”
Bottom line: If you can’t communicate your work in a way that makes sense to people beyond your team, you should start working on that now. And during the machine learning interview process, make sure you articulate the ways you do communicate internally and externally (including how they differ).
7. Explain the steps in making a decision tree.
Wieczorek also asks data science job applicants this question every time she interviews. While you may not be interviewing for a data science job but rather something more entry-level, you should still expect a machine learning job interview question that’s centered on how you make decisions or what processes you use. Why?
“It helps me understand their decision-making process,” says Wieczorek. How you tackle this answer—and whether you can explain and outline it clearly—will give interviewers insight into how you’d work within (or run) a team.
Bottom line: When it comes to machine learning and data science, you need to know how to make educated decisions. Think about the ways you’ve done that most successfully in the past and be prepared to use them to illustrate your process.
8. What is the goal of A/B testing?
Wieczorek throws this question in the mix for what she describes as a “general knowledge test.” For this and all machine learning and data science interviews, make sure you know the best practices in your field. You will get asked about them.
Bottom line: Know the general areas of machine learning and/or data science and what the current standards are but also why they’re standards.
9. Why does data cleaning play a vital role in analysis?
Last but not least, Wieczorek gives candidates what she calls a “technical knowledge test” like this question. Max Babych of SpdLoad seconds this approach, explaining (for data science roles, at least), “most of the interview questions are in mathematics.” Make sure you’re comfortable with the more technical skills you’ve learned—especially the ones listed on the job description—and expect detailed questions around them.
Bottom line: Yes, technical knowledge matters, but if you’ve got the chops (like Python knowledge), you’ll do just fine.
10. Give me some context around what this set of data was created in, as part of the analytical process.
Esquivel likes this question because it forces candidates to think about the flaws in data—a crucial part of the skill set. Says Esquivel:
“It’s easy to point out and understand that the measurable, structural attributes of the data we’re working with often correlate with high model performance metrics. However, it’s much less intuitive to explain how the social and cultural context a dataset was created in can lead to models that replicate the biases of, or rely on inferences exclusive to, that context, leading to negative feedback loops, incorrect sample weighting and scoring, and unexpected or even invalid results when applied to a different or larger context.”
Bottom line: When answering questions around the analytical process, don’t just think about the positive results, consider the negative implications (or at least the possible flaws) as well.
11. Is machine learning a science or an art?
Esquivel came across this curveball question when he sat in on a series of interviews for a machine learning role. Here’s why it’s sneaky, he says:
“The interviewers were not so much interested in hearing how it was a mix of both, but on using the question as a springboard towards a discussion on the importance of following rigorous scientific practices, and the need for creativity as part of the analytical process; interpreting the data, formulating ways in which it might be useful, turning model results into actionable insights, etc.”
Bottom line: You need to understand how essential both creativity and scientific practices are within your role—and to articulate how they work together. This is also an excellent opportunity to talk about whether you’re more creative or scientific, and the ways you’ve worked to offset that imbalance to make yourself a more well-rounded machine learning engineer.
12. When is it better to use TensorFlow [or another library, technology, or approach]?
When it comes to machine learning (or even data science interviews for those looking to land jobs in the C-Suite), expect questions that test how up-to-date you are on current technologies and trends. This includes questions related to Python libraries like TensorFlow. Babych, who once had to hire seven data science PhDs for a three-month project (meaning he conducted more than 100 interviews), loves sprinkling questions like this between the more mathematical ones. “I really like questions about modern technologies and approaches.”
Bottom line: Don’t be afraid to have opinions as long as you can back them up.
13. How would you remove rain from a short video?
Babych also likes “quick questions on logic” that keep interviewees on their toes. This question is a trick, he says, because “a specialist could never do this, [so] this is an exclusive question for quick logic.“
Bottom line: Expect trick questions. The best way to deal with them is to listen carefully and don’t rush your answers. Taking time to process and consider isn’t a bad sign—it proves you take your approach seriously.
14. Give an example of a problem you solved (or tried to solve) with machine learning.
Stormoen (the founder of Mobibi mentioned above) loves this question for its simplicity. His advice? “Make it personal—[show your] passion, self-starter [approach], creative thinking.”
Bottom line: Yes, even machine learning or data-driven jobs require passion. Show yours off a little.
15. Expect technical questions around frameworks and machine learning models, plus some conceptual questions, too.
Stormoen says you should “know your frameworks and common ML models, e.g. PCA, Regression, and Clustering.” You should also be able to “demonstrate an understanding of common ML stacks at the cloud providers (AWS, Azure, or GCloud) for example with AWS – S3 buckets or RDS + lambda + AWS ML, etc.”
Or take this advice from Allen Lu, a former machine learning intern at Google, who outlined the machine learning job interview questions he tackled to land the internship in a recent Quora post:
“Common questions will focus on basic machine learning topics like logistic regression, SVM, Naive Bayes, etc. You’ll also likely be asked about fundamentals for neural networks, such as fully-connected layers, activation functions, and the pros/cons of deep learning (more hidden layers).”
Google didn’t stop with the technical questions, though. (Why would they?) Much like the decision-making related questions we mentioned earlier, Google also likes to get a little conceptual. Says Lu, “[Y]ou can expect some conceptual questions, such as the bias-variance trade-off, different loss functions, overfitting/regularization, etc.”
Bottom line: Again, the technical interview questions are going to vary, but the best way to prepare for them is to review all the basic machine learning topics you’ve learned so you feel comfortable talking about them.
16. Curveball: The German Tank Problem
Tip: Read this “question” carefully. It’s one that Pete Sosnowski, the Co-Founder and a Head of HR at Zety often uses to screen candidates, and it’s a little… complex. Here’s how it works (Sosnowski bolds the key steps below):
“How many tanks did Germans produce?” asked the Allies during WWII. And the question would be left without an answer, if not for the fact that Germans (knowing their passion for “ordnung”) liked to number their tanks sequentially when they were leaving the factory. So now we have an unknown number of tanks which are sequentially numbered from 1 to N. Take a random sample of the tanks spotted (no 12, 37, 52…) and observe their sequence numbers; then use your data science skills to estimate the N (total number of tanks) from these observed numbers.”
“The answer to this question won’t help the Allied forces at this stage, but it will definitely let us see the thinking process of the machine learning applicant,” says Sosnowski, “It’s a technical-task-type question, but the way it’s approached (the method chosen to solve it and the solution itself) lets us fish out the candidates with the biggest potential.”
Bottom line: Breathe, baby, breathe. Sometimes there’s no “right” answer—the question is designed to show the way you think. So let that brain buzz and just do your best.
Good luck with your interviews! And don’t forget to check out our other tech job interview question round-ups if you’re in the market for a tech industry gig.
Kit Warchol is the Head of Content for Skillcrush and writes for magazines and sites including Fast Company, Entrepreneur, Girlboss, and others in her spare time. After teaching herself to code at the height of the recession (heyo, 2009), she worked as a web designer at various tech startups, then took a Senior Project Development role at the University of Southern California before diving back into writing full-time. Before joining us, she served as the Editorial Director of Career Contessa, a career advice site for women.