Get Started With Machine Learning: It’s Not Too Late

Source: Deep Learning on Medium


Robbie Allen and Dr. Larry Carin speaking at Infinia ML’s Rethic Symposium at UNC Kenan-Flagler (November 2018)

As machine learning and artificial intelligence become more commonplace in our daily lives, businesses are increasingly recognizing the possibilities these technologies hold.

For years, news stories have been discussing the pros and cons of artificial intelligence and machine learning. Gradually, companies in nearly every industry are seeing how these technologies can make their operations more efficient, remove costs from their processes, and even uncover new opportunities.

But companies that haven’t yet explored artificial intelligence and machine learning might be concerned that they’re too late and will have to play catch-up. Or, they may see the potential but have no idea where to start.

Those companies can be assured that it’s not too late to implement artificial intelligence and machine learning, and there is a clear path to getting started.

Just the Beginning

Machine learning has been around for decades; Arthur Samuel, who created one of the world’s first self-learning computer programs, coined the term in 1959. Its ascension to business prominence has been more recent, and it has been at the peak of Gartner’s Hype Cycle for the past four years.

As fast as technology moves, it’s natural to think that if you haven’t adopted machine learning yet, you’re behind the times. When our Chief Scientist Larry Carin and I started Infinia ML back in 2017, we ourselves worried that we might be starting too late. But our experience working with clients shows that nothing could be further from the truth.

To understand why, it’s helpful to draw a parallel to the early days of electricity. Thomas Edison perfected his light bulb in 1879, but it would take more than a generation for every home and business across America to glow with electric light.

That’s because prior to 1879, the ability to generate and distribute electrical power on a large scale didn’t exist. The first major power station in America began operating in 1882, but it would be decades before electricity was widely available in every corner of the country. It took time to create the electrical infrastructure.

That’s where we are today with machine learning. Data scientists are constantly creating new ways businesses can harness their data to optimize their operations, but to a very large degree, the infrastructure is still not in place.

In terms of machine learning, the infrastructure consists of a data science culture within companies. It’s more than simply having some data. In companies with a data science culture, capturing and learning from data is a part of their DNA. It’s an important part of what they do, just like finance and human resources.

While some companies are starting to get better at creating a data science culture, very few have truly mastered it. If your data science culture is still emerging, you haven’t missed out. Machine learning may have been around awhile, but we are still only in the beginning stages of realizing its full potential.

Where to Start

Once companies understand they still have the opportunity to implement machine learning, and they see how ML can benefit their business, the next logical question is where to start.

Here are three recommendations:

Focus on the data you already have.

Data is the lifeblood of machine learning. The reliable data that’s already being captured at scale can help define the range of problems that you can attack.

Pick problems with a good chance of success.

When you’re starting to implement machine learning, it’s important to get some wins. Avoid being too ambitious by trying to solve your organization’s biggest, most expensive problems. That’s a formula for spending money with little or no return.

Instead, focus first on problems with achievable solutions that can deliver immediate business impact. This will lay the foundation for attacking bigger problems in the future.

Don’t overlook old tools.

Recently, we solved a problem for a client using factor analysis, a technology developed decades ago. Despite its age, it’s a technology that’s far from outdated.

Sometimes older tools are still ideally suited to today’s problems, but newer data scientists may not be aware of them. They’ve been trained on the latest, cutting-edge deep learning techniques.

Advanced techniques like deep learning are a massive hammer. But not every problem is a nail. Be sure to bring in people with depth and experience who can evaluate your problem and recommend the right techniques, regardless of their age.

The Future of Machine Learning

The academic field of machine learning is moving forward rapidly, but we are starting to reach the limit for big ideas. What’s most exciting about the coming years may not be what gets published, but instead what makes the greatest business impact. In five years, we expect AI and ML will be integral to almost every large and serious company.

As companies demand better solutions, machine learning tools will become better, more efficient, and less expensive.

Simpler, too. Machine learning has grown up in academic environments where models have been layered upon one another, creating complex solutions to achieve relatively small gains. In the coming years, we may see some of that complexity get stripped away.


Machine learning is moving out of the ivory tower and into the real world, and it’s an exciting time to be in business. The time to embrace the promise of machine learning is not in the past — it’s now.

You just need to get started.