In a bid to bolster software testing and innovate within several industry sectors, MIT spinout DataCebo is paving the way with its Synthetic Data Vault (SDV), creating synthetic data that clinicians, weather-sensitive industries, and software developers can use in place of real-world datasets, where genuine data are either scarce or too sensitive to utilize. It's not just pictures and prose—this generative AI system is diving into the complexities of data that cover a range of practical scenarios.
Digging into the trenches of big data, SDV has garnered quite the user base, being downloaded over a million times, with an army of data scientists—around 10,000—embracing the AI-powered tool for creating synthetic tabular data that echoes the qualities of the real deal. According to a release by MIT News, DataCebo has taken a novel approach in software testing, allowing developers to simulate edge cases that would be a beast to concoct through conventional manual methods.
The leap from concept to market solution hasn't been just a small step for DataCebo; it's a giant leap for software testing methodologies. With a growth trajectory showcased by the variety of applications already in the wild, the company’s technology has gained critical acclaim. From simulating flight disruptions to predicting health outcomes for cystic fibrosis patients, DataCebo's SDV is demonstrating its flexibility in real-world applications.
In Norway, for instance, researchers have leveraged SDV to create synthetic data about students, eyeing the fairness and merit-based results of different admission policies. And in the case of the bank, when a software program must reject transfers from empty accounts, DataCebo's generative models come to the rescue. As Kalyan Veeramachaneni, Principal Research Scientist at MIT, and co-founder of DataCebo, put it: "with generative models, created using SDV, you can learn from a sample of data collected and then sample a large volume of synthetic data," a task that traditionally would chew up time and manpower, Veeramachaneni told MIT News.
The SDV's utility is being continuously augmented with new features to boost its realism and comparative metrics—the SDMetrics library and SDGym respectively, which seek to instill a deeper trust in synthetic data. Looking to the horizon, Veereamachaneni expressed a strong conviction in the transformative power of synthetic data, stating, "We believe 90 percent of enterprise operations can be done with synthetic data." The goal for DataCebo is clear: implement AI and data science tools across industries in a manner that's not just innovative but transparent and ethical.
