Cookie Consent

By clicking “Accept”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts. View our Privacy Policy for more information.

Work With Your Data, Not Against It

March 7, 2022
Ken Stibler

Ken Stibler

Data is like oil as pundits never tire of saying; it has massive potential to power value-generating activities. And like oil, data is worthless if it’s stuck in the ground. Modern “data-driven” enterprises produce vast swaths of data but struggle to access the insights that machine learning efforts promise, due in large part to challenges in accessing and using that data.  Unlocking access to this data is key to providing more business value and innovation, including granular understanding of customers and their behavior, substantial cost savings, and better decisions through ML.  Devron’s machine learning platform allows Data Scientists to train models on siloed, private datasets unlocking deeper insights.

IDC says 68% of all data collected by organizations goes unused, which translates into less accurate models, low ROI for ML initiatives, and missed opportunities. The graveyard of failed COVID healthcare models, inequities of AI-enabled hiring, and costs of large investments on incomplete predictions emphasize the problems of stranded data. Such challenges aren’t new though, with data always being AI’s biggest impediment and silos, bias, and restrictions putting the pinnacle of machine learning value out of reach for the modern enterprise.

Proposed solutions have emerged to try and unlock some of the data barred away from use. Differential privacy (adding noise to the data to decrease privacy concerns), has gained increasing interest. Synthetic data (generating artificial records) has also become a popular method to increase the quantity of information available. However, each option just replaces its respective problems with others. Differential privacy reduces data quality and impacts model accuracy. Synthetic data can actually magnify bias while creating unrealistic output.

While differential privacy and synthetic data might partially address the issues, the ability to train models directly on siloed/distributed datasets enables enterprises to leverage their locked data for high-quality models while avoiding regulatory or bureaucratic restrictions.  Enter Devron.   Devron has created a platform enabling data scientists to train models at the point of data collection and centralize only model weights and parameters rather than the data itself, leveraging technologies such as federated learning.  

DiagramDescription automatically generated

Source: Open Data Watch

With Devron’s platform, each data silo becomes a training set, and trapped insights are sprung with the added benefit of context. Federated model training therefore frees otherwise stranded data to fuel AI, just like once inaccessible oil powers today’s industry. Armed with this additional data, machine learning models perform better, meaning AI keeps its promise of higher predictive accuracy to lower risk and improve customer engagement.

See how Devron can provide better insight for your organization

Request a Demo