The Hard Questions of Hiring For Machine Learning

I’ve been thinking a lot about hiring for the machine learning specialization lately. It’s no surprise. New data is emerging almost daily about the rise of machine learning, artificial intelligence and deep learning in software design. Just recently, IDC reported that spending on cognitive and artificial intelligence (AI) systems is set to accelerate well beyond original forecasts by more than 300% in the next five years. Even a survey at my company revealed that almost two-third of enterprises are experimenting with AI.

This means we’re always on the lookout for machine learning talent to work on our service-centric AIOps platform, and it’s like panning for gold. The New York Times reported that there are about 10,000 people in the world with the skills to handle the hardest problems in AI. Of course, we’re not just looking for one of these unicorns. We need teams of them to work on building solutions with neural network architecture, Naive Bayes Classifications, and Singular Value Decomposition. That means we need experts in data science, who can use data to validate models, and engineering, who can code the mathematics into the software. Simple, right? Not really. Here’s how we attack this problem fundamentally, with a few of our tried-and-true interview questions that help us find the intelligent minds behind artificial intelligence.

Data: The Ghost In the Machine

To start with, it’s critical for our candidates to understand the value of data in building artificial intelligence. In fact, as technology gets more complex with the rise of discrete, ephemeral workloads like containers, microservices, cloud instances and serverless computing, the amount of data that can potentially be fed into an artificial intelligence solution is multiplying exponentially. That means a scalable solution must be able to handle large quantities of data quickly and accurately.

Data is what underlies our platform and it’s at the core of any algorithm built to solve a problem, so a deep understanding of statistical methods and data manipulation is key. Ideally, a candidate would come to us with an advanced degree in these fundamentals, but in the absence of that, we ask some questions to get at the heart of their allegiance to data, including:

1. How do you prepare data for machine learning and how do you convert different objects into something the model can learn from?
2. How do you tune the hyper-parameters to get the best model?
3. How do you take into account overfitting to your dataset and how do false negatives and false positives affect results?

This way, we can start to understand how a candidate would manipulate data inputs into something an algorithm can find useful. Usable data must be comparable, with limited false instances to power accurate inferences.

Machine Learning: The Code of Mathematics

Machine learning engineers are the ultimate data scientists who code. The right ones bring the best of software engineering, applied mathematics and the business focus of product management into a single role. They’re difficult to find, but the right ones can transform a project and make it competitive in a crowded space of AI-enabled solutions.

Because of this hybrid perspective, we ask our machine learning engineering candidates to solve a few scenarios at the outset of our interview process. We also ask them about their comfort with fundamental tools that have emerged in the space. Together, these sets of questions help us identify engineers who know the models, have worked with the tools, and understand the business. Some of these include:

1. What machine learning models have you worked with? Why?
2. Do you have experience with Apache Spark or some public machine learning libraries like Tensorflow and Gluon?
3. What data visualization libraries do you use? What are your favorite data visualization tools?

Attitude: The Great Unknown

Finally, we also hire for sheer problem-solving muscle. A machine learning engineer needs to believe that data, math and code can solve some of the hardest problems in business with accurate, scalable and fast results. They need to be able to see solutions that are years away, and be able to visualize a future where many routine tasks can be handled unsupervised.

(Dean Drobot/Shutterstock)

In this stage of the interview, we ask them to paint us a picture and envision scenarios, or create case studies for the candidate to solve. If they bring case studies from previous positions, that’s even better. Some of our questions would include:

1. What are the steps from start to finish for a machine learning pipeline?
2. How would you handle an imbalanced dataset?
3. When should you use classification over regressions?
4. How do you handle missing or corrupted data in a dataset?

Here, we can determine how a candidate tackles these problems in real-time.

Hiring For the Future

There have always been stories of the difficult hiring processes of some of technology’s biggest brands, like Google and Facebook. Today, those hiring processes are almost quaint in the world of machine learning. Engineers who build for the future are also physicists, working with huge numbers on an astronomical scale. They’re researchers, looking for mathematical models that can be applied to real-world scenarios. They’re technologists, with a background in the latest and greatest in coding, DevOps and cloud-native tools. And they’re business leaders, with an eye on building solutions to serve customers today and tomorrow.

If every company is now a software company, that means that any company can be an artificial intelligence company. And the race is on for talent. Nobody said that building these solutions was easy. But if you assemble the right team, the possibilities are endless.

About the author: Bhanu Singh is vice president of product development and cloud operations at OpsRamp, a provider of service-centric AI ops solutions. An accomplished and decisive leader in the software technology industry, Bhanu is an expert in large-scale global engineering management, software development life-cycle, operational process optimization, transformation projects, global talent development, and customer engagement.

Related Items:

Why Data Scientists Love the Law of Supply and Demand

Improving Your Odds with Data Science Hiring

How To Build a Data Science Team Now