Skip to Main Content

How a Great Company Gets Data for Machine Learning

group working on computers

You’ve probably heard about machine learning. Even if you are not completely sure of what it’s about or what it’s for, you kinda know that machine learning is taking over the world.

Why? Because it’s powerful, simple, and effective. However, this is not a story about the magnitude of machine learning itself. This is a story about how great companies rally around a common purpose to succeed.

What is machine learning? Simply put, it’s just computer programs that are capable of solving problems without a specific formula or recipe to do so. Machines are using examples to learn over time. In a way, it’s very similar to the way we learn as humans – by experience.

Since machine learning relies on examples (data) in order to build accurate models to solve problems, the collection of that data is crucial to the successful development of machine learning systems.

At Accusoft, we are actively using machine learning to improve our products and gain a competitive edge. However, we experienced the same roadblock as many companies face when beginning this process. Our existing data was not sufficient to get our system to be as good as it could be.

When we realized the issue, we knew we needed to think outside of the box. In a nutshell, we needed thousands of additional data files that required thousands of hours of manual work, but we didn’t have enough resources to make it happen within a reasonable timeframe.

group working on computers

So…What did we do? We asked for help.

We sent an email to the entire company explaining that our product group needed everyone’s help if we wanted to succeed at creating data for our machine learning system. We included a volunteer sign-up sheet that allowed anyone in the company to sign up for half-hour increments for the next four days.

The response? On the first day, people all over the company signed up for 65 half-hour slots.

You would think that participation decreased after the initial hype…right? Quite the opposite. Participation actually increased almost 10 percent for the last two days.

Every division of the company participated. Human resources, accounting, sales, marketing, technical support, IT, product management, and engineering…It was absolutely incredible. I’ve never seen people so excited to help.

Participation was impressive at all levels. Our CFO, the VP of Engineering, VP of Product Management, engineering directors, engineering managers, and more. The CEO even participated!

Why go through all this trouble, you might ask? Couldn’t you hire some contractors to do the same thing? Perhaps. However, we accomplished in four days what we planned to do in a few months of work with contractors.

In addition, there was zero overhead, and we avoided a long drawn-out planning phase. More importantly, we were never going to get the same level of satisfaction, ownership, and sense of community that tackling this challenge as a company offered.

Great businesses have exceptional employees that act like a family. Accusoft pulled together as a family, conquering one of the greatest challenges we were facing at the time. We worked as a unit with one goal. Because that’s what you do when someone in your family needs help. You lend a hand.

 


 

Josh Candamo, Director of SDKs

Josh Candamo, Director of SDKs

Joshua Candamo, PhD, Development Director for the SDK product group, oversees the development and maintenance of 22 of Accusoft SDK imaging products. He believes that your most valuable intellectual property has nothing to do with patents or technology, but everything to do with your people. He is passionate about team building and creating the right corporate culture to develop amazing software products. Josh joined Accusoft in 2015 after a career in software development that included technology leadership, entrepreneurship, consulting, and both back-end and front-end development. He holds a PhD degree in Computer Science from the University of South Florida, specializing in pattern recognition and image processing.