ML Deployments Slowed by IT, Resource Issues
The bigger the company, the more likely its success in deploying machine learning models, according to a new survey that also found customer loyalty programs and cost-saving initiatives are the most popular enterprise use cases.
Algorithmia, the machine learning deployment specialist, reported this week that, despite heavy investment in model development, data science teams continue to be sidetracked by infrastructure and deployment issues. The result is that they are spending on less than one-quarter of their time on model training and refinement.
“Managing machine learning models remains a challenge,” the survey found. Deploying “models is seen as the last step to ROI. Without an enterprise platform to help, these companies are missing out on the rewards of machine learning.”
As with most vendor surveys, the Seattle-based startup points to the results as evidence supporting its tools. In Algorithmia’s case, it’s an infrastructure category it refers to as the “AI layer” of the protocol stack for managing computing resources, automating the deployment of machine learning models across companies. Facebook and Google (NASDAQ: GOOGL) have released similar AI management tools.
“Larger companies have more machine learning use-cases in production than smaller companies,” said Algorithmia CEO Diego Oppenheimer. “But across the board, all companies are getting smarter about where and how to apply ML technology.”
The startup’s AI Layer is designed to accelerate the development and deployment of machine learning models. The company claims its framework fills a gap in the current AI infrastructure that prevents machine learning investments from reaching production.
Despite lingering infrastructure and resource constraints, Algorithmia also said it expects more machine learning models to shift to production in 2019 as data scientists make use of new tools for deploying and managing their models.
With large enterprises making heavy investments in data science as a way to streamline operations, nearly half of survey respondents cited “cost savings” as a primary machine learning use case. “Large enterprise companies have an advantage when it comes to machine learning because they have access to more data, can continue to invest in big R&D efforts and have many problems that machine learning technology can solve cost-effectively,” Oppenheimer added.
Meanwhile, 30 percent of those polled reported issues in supporting different programming languages and training frameworks. These distinct languages and frameworks create another layer of complexity in model development since they must be pipelined together for a specific task. The survey notes that companies like Facebook (NASDAQ: FB) and Uber have built internal tools to address these issues, including FBLearner and Michelangelo.
Algorithmia said Tuesday (Oct. 16) it surveyed more than 500 data science and machine learning practitioners.