H2O.ai: $72.5M Funding Round for AI Democratization 

Ambati points to customer success as evidence that his Mountain View, CA, company is making a difference in the AI initiatives of its customers. Among the companies that have adopted H2O.ai software are Aetna, Booking.com, Capital One, Comcast, Commonwealth Bank of Australia, Franklin Templeton, Hitachi, MarketAxess, Nationwide Insurance, PwC, Tokio Marine, Wells Fargo, and Walgreens.

“H2O.ai is democratizing AI and powering the imagination of every entrepreneur and business globally,” Ambati declared in a press release. “Our customers are unlocking discovery in every sphere and walk of life and challenging the dominance of technology giants. This will be fun.”

The Goldman Sachs-led Series D round brings another $72.5 million investment to H2O.ai, bringing the company’s total funding to $147 million. “We’re thrilled to partner with the H2O.ai team on their mission to democratize artificial intelligence,” said Jade Mandel, vice president of the Principal Strategic Investments Group at Goldman Sachs. Mandel has been added to H2O.ai’s board.

H2O.ai gained the big data spotlight years ago with the release of its open source machine learning library, which has become one of the most widely used libraries among data scientists. The package contains a variety of machine learning algorithms, including a generalized linear model (GLM) that’s widely considered to be one of the highest performing regression algorithms around. In 2018, the company claimed that 18,000 companies were using the core H2O library, spanning more than 200,000 end users across all industries.

The company shifted gears with the late 2017 beta release of Driverless AI, a proprietary AutoML offering aimed at enterprise adopters of AI. In addition to data science-focused tasks like model selection and hyperparameter tuning for traditional machine learning and deep learning models, Driverless AI automates more engineering-focused aspects of the AI pipeline, including data preparation, creating test and validation data set, comparing models, and deploying the models to production.

(H2O.ai originally described Driverless AI as like having a “Kaggle grandmaster in a box.” Interestingly, as part of its global expansion, the company boasts that it actually employs 10% of the world’s Kaggle Grandmasters.)

The company began selling subscriptions to Driverless AI in early 2018, and since then adoption has ratcheted up. Today the company announced a new release of Driverless AI that brings new features in four key areas, including:

• A “recipes” features that will expand customization options;
• The availability of 100 open source DriverlessAI recipes;
• Model administration and collaboration features for model deployment;
• Explainable AI capabilities to check for fairness and bias.

The new recipes feature is designed to help data science teams manage how they tweak certain elements, including the models themselves as well as transformers and scorers. Combined with the A/B testing features of Driverless AI, it will help teams build and identify winning approaches.

Yan Yang, VP of data science at Deserve, says he’s pleased with the new recipe features in Driveless AI. “We can be more creative in how we evaluate and serve those new to credit with the ability to customize and extend the platform to meet our unique needs,” Yang says in an H2O.ai press release.

The new model management and collaboration features, meanwhile, will help teams build models, tag, and version them appropriately to make it easier for DevOps and IT to deploy them to production. Driverless AI also sports a new module designed to monitor deployed models along certain metrics, and to alert when models drift or degrade. Integrated A/B testing will also help alert administrators when it’s time to recalibrate and retrain the models, H2O.ai says.

H2O.ai also added a host of various techniques and methodologies for explaining the results of its models. Including K-LIME, Shapley, variable importance, decision tree, and partial dependence views, the company says. It also includes a new feature that will automatically check whether a model “produces adverse outcomes for different demographic groups even if those features were not included in the original model,” the company says.

These explainability features will be well scrutinized, particularly among companies that are debating whether they can adopt AI while keeping an eye toward ethics. One person who’s looking forward to the new feature is Agus Sudjianto, the executive vice president and head of Corporate Model Risk at Wells Fargo.

“Decisions made by models must not only be sound but also must be fair,” Sudjianto says in a press release. “The team at H2O.ai has tackled this with machine learning interpretability and, disparate impact analysis to detect bias and fairness. H2O.ai by far has the most sophisticated and complete tools to address these critical requirements for data scientists today.”

This article originally appeared in sister publication Datanami.