Machine learning (ML) seems to work so well for big tech companies while so many businesses outside of Silicon Valley have yet to fully implement ML to its fullest. I recently visited with Monte Zweben, CEO of Splice Machine, creator of an ML engine that creates, deploys, and manages ML models, and I asked him why.
SEE: TechRepublic Premium editorial calendar: IT policies, checklists, toolkits, and research for download (TechRepublic Premium)
“It is incredibly complex to build a data architecture from scratch that supports high-quality machine learning,” he said. “It takes a significant investment of time, money, and talent to do so, and big tech companies invest highly in talent that is on the bleeding edge of the data innovation cycle. They hire software engineers and computer scientists with postgraduate degrees who are able to duct-tape together components from cloud vendors and open source communities, which they tweak until they find a good fit. These companies have the added advantage of building their data on modern, cloud-based IT infrastructures that integrate easily with machine learning technology.”
Just to grasp the implications of these statements is a tall and almost unachievable order for everyday companies that don’t have the resources to invest in such top-tier talent. Small- and medium-sized businesses also have the additional challenges of upgrading enterprise legacy system infrastructures for ML. This is a daunting project in itself that many high tech companies don’t have to concern themselves with.
As a result, it makes sense for companies not in high tech to look for pre-configured machine learning solutions if they want to incorporate machine learning into their systems. However, before companies can consider purchasing pre-configured ML solutions, they have to understand what ML is, what it can do for them, and whether ML is worth the investment.
What is machine learning?
Machine learning is a subset of artificial intelligence (AI). It uses a set of predefined algorithms to detect patterns in data and to draw conclusions that enable the AI to “learn” from the data that it’s processing.
To facilitate the learning process, ML looks at “features” that summarize raw data at a higher level of understanding.
“For example, the raw data for temperature readings might be expressed as 27 degrees or 90 degrees,” Zweben said. “Temperature data in this form might provide less useful information than if it were summarized by an algorithm at a higher level, such as ‘cold,’ ‘warm,’ or ‘hot.’
In this way, it becomes easier for machine learning to work with and to ‘learn’ from these higher-level, summarized features, instead of trying to interpret raw numbers such as 27 degrees.”
Data feature stores help reduce time
Machine learning vendors now offer higher-level data feature stores that can reduce the amount of time data scientists have to do the same thing as they work through raw data and try to summarize it at higher levels for use in AI and machine learning. The pre-configured data feature stores save valuable data scientists prep time and enables machine learning technology to be more readily used by business end users, who no longer have to wait for data to be transformed into something usable that makes sense for the business. When this happens, more people observe ML in action for the business and begin to understand its value.
SEE: Natural language processing: A cheat sheet (TechRepublic)
“There are a number of companies that are working to take the complexity out of data plumbing so machine learning can be democratized and easily integrated into business,” Zweben said. “But for machine learning to work, there needs to be an integrated approach to incorporating ML into companies, as data scientists and businesspeople will perform best when they work together to achieve common goals. Data scientists should be present in business meetings and communicate what ML can accomplish and what has been achieved with it, while minimizing the technical how-to. On the flip side, having data scientists in the business environment will give them a better idea of what to optimize so their models can better reflect the business domain. Overall, a culture of experimentation that integrates machine learning seamlessly into the business will be most successful.”