Upgraded platform also includes significant enhancements to MLOps, Automated Time Series, and Visual AI capabilities
DataRobot, the leading enterprise AI platform, today introduced automatic Bias & Fairness Testing to identify bias in models with protected features such as gender and ethnicity, then provide guidance to resolve upstream issues and prevent bias from reoccurring in the future. The company also unveiled significant enhancements to the platform’s key capabilities—including MLOps, Automated Time Series, and Visual AI—to further accelerate the value organisations achieve from their AI models.
According to research conducted by DataRobot, 42% of organisations are “very” to “extremely” concerned about AI bias, citing compromised brand reputation and loss of customer trust as the two largest causes of concern. To prevent this from happening, DataRobot has made an industry-leading commitment to ethical, fair, and explainable AI solutions—building out an AI trust team and providing robust capabilities to help customers ensure models deployed on the DataRobot platform do not exhibit bias.
In support of that commitment, the company today released the Bias & Fairness Testing feature to automatically identify model bias and determine its source. With Bias & Fairness Testing, users can define protected dataset features, and through a guided workflow, choose the most appropriate fairness metric to fit their specific use case. Once models are built, DataRobot surfaces visual insights to illustrate the results. If bias is identified, the Cross-Class Data Disparity tool surfaces the root cause and identifies mitigation steps for future models.
“Customers consistently turn to DataRobot for our focus on usability, value, and trustworthiness of AI. With AI being relied upon to empower organisations to improve mission-critical decision making, it’s become even more imperative that they think critically about the ethical implications of the data they’re using to train their models, especially to prevent inadvertent bias,” said Nenshad Bardoliwalla, SVP of Product. “Our new Bias & Fairness Testing capabilities further strengthen our customers’ ability to build trustworthy, explainable AI models that generate real business value.”
While most organisations struggle to derive real business impact from AI, DataRobot’s customers achieve a 514% return on investment—in large part due to the robustness of the DataRobot platform which now includes these new features:
- MLOps Portable Prediction Servers – Portable Prediction Servers allow MLOps users to integrate any DataRobot model into the pipelines and applications they have already built outside of their DataRobot environment. Portable Prediction Servers are dockerised containers, making it much easier for IT and DevOps to deploy, monitor, and manage models on a customer’s platform of choice.
- Time Series Feature Lineage – A key strength of DataRobot’s Automated Time Series product is the specialised feature engineering it performs to automatically generate expert new features for time series models. Now, DataRobot provides a visualisation showing the full lineage of every new feature generated, helping users understand how they were derived and providing even more governance and traceability for time series models.
- Automated Deep Learning in Time Series – In addition to feature lineage, DataRobot has also added new cutting-edge deep learning approaches to forecasting that will run automatically as part of the Autopilot process. As a result, users always have the widest variety of traditional and emerging time-series modelling techniques for every project.
- Visual AI Smart Autopilot – Visual AI’s Autopilot will now automatically test a variety of deep neural networks, selecting the most appropriate one based on the specifics of the use case. In addition, DataRobot will automatically compare alternative neural networks on the top leaderboard models to maximise accuracy and minimise the need for manual tuning.