Algorithmia Adds Data Governance Tools
Data governance and a growing list of compliance rules remain key considerations for managing machine learning models in production. Vendors are responding to the growing risks with compliance frameworks as they flesh out their enterprise offerings.
Among them is Algorithmia Inc., which this week released new reporting and risk-assessment tools designed to govern the use of machine learning models in production. The MLOps vendor said the management tools respond to growing concerns about data governance and security as they accelerate the launch of production workloads.
An internal company survey also revealed that two-thirds of respondents report they must comply with multiple regulations when deploying machine learning models in production.
Risk management has traditionally focused on testing and validating models prior to deployment. As machine learning deployment accelerates, new risks are emerging with production models. “Operational risk is now the most significant analytics risk,” Algorithmia noted this week in releasing its new governance features.
The Seattle-based company’s reporting and governance tools include audit reports and logs for reviewing model results, model-change histories, a record of data errors or previous model failures along with the actions taken to correct errors.
Also included are cost and usage data on infrastructure, computing and storage resource consumption along with algorithm usage reports.
The company asserts its framework goes beyond current data governance steps it describes as a “patchwork of disparate tools and manual processes.”
Added Algorithmia CEO Diego Oppenheimer: “Regulations are undefined and a changing and ambiguous regulatory landscape leads to uncertainty and the need for companies to invest significant resources to maintain compliance.”
The new data governance tools would allow early adopters of machine learning models to manage them in production within their existing data governance, compliance and risk assessment frameworks. The enterprise tools reflect the MLOps specialist’s early focus on model deployment, management and risk assessment.
“Governance is by far the top challenge for AI/ML deployment, with more than half of all organizations ranking it as a concern,” the company reported in its annual survey of trends in machine learning. Those concerns focus primarily on minimizing risk once models have been developed and deployed.
As the number of production models grow, early adopters said the challenge in leading use cases such as automating business processes is scaling MLOps tools for data governance and security.