AT&T and H2O.ai Team Up on Feature Store

Telecom giant AT&T and AI startup H2O.ai have teamed up to co-develop a feature store, which the companies announced today. The jointly developed H2O AI Feature Store will be available for any company to help better manage their data science development and production efforts.

The feature store is among the hottest segments in the data science market at the moment, thanks to the efficiencies they can bring to data science organizations. The key advantage that the feature store brings to its user is the ability to re-use the various data features that data scientists have previously used in a machine learning application.

Instead of starting from scratch to figure out which data features or metrics are the most important or predictive for a given machine learning project, the data scientist or ML engineer can look in the feature store to see what features worked in the past, and then re-use those features for the new application.

AT&T worked with H2O.ai and its machine learning and AI software in the past, and its data and IT executives decided to go into business with the Mountain View, California-based in developing the feature store, according to Mark Austin, vice president of data science in AT&T Chief Data Office.

“H2O.ai was a natural fit to co-develop a feature store with AT&T,” Austin told Datanami via email. “We were impressed by H2O.ai’s capabilities in terms of building powerful machine learning models and deploying them, and there was a need for a solution to capture the predictive features, regardless of the pipeline that created them, in order to reuse them for different use cases, and serve them up for real-time machine learning scoring.”

AT&T is no stranger to machine learning. In an effort to streamline machine learning development across its enterprise, the company worked with H2O.ai develop the feature store, which it will use with various machine learning projects, Austin said. “The feature store solved a product niche for H2O.ai and helped us speed development through feature reuse across many teams and use cases at AT&T.”

The reason why feature stores are so hot is due to the capability to reuse and repurpose data engineering tools in complex and expensive programs, according to AT&T Chief Data Officer Andy Markus.

“These storehouses are vital not only to our own work, but to other businesses, as well,” Markus states in a press release. “With our expertise in managing and analyzing huge data flows, combined with H2O.ai’s deep AI expertise, we understand what business customers are looking for in this space and our Feature Store offering meets this need.”

The H2O/AT&T feature store, which can be used for training machine learning models as well as using them for production inference workloads, boasts pre-built integration with popular AI environments and frameworks, including Snowflake, Databricks, H2O Sparkling Water, and Apache Spark, according to the product’s webpage.

“When features are written to the H2O AI Feature Store, data scientists can specify over 40 metadata attributes, tags, and the set of features that need to be available for real-time applications,” it states. “H2O AI Feature Store uses built-in AI to automatically recommend new features, identify bias, and create feature insights.”

The feature store enables data scientists to search the feature store to find the features the need for their models, which helps to build more accurate and robust models faster, H2O.ai says. For inference workloads, the feature store delivers results with sub-millisecond latency, H2O.ai says.

It also sports a concept called feature rank, says Sri Ambati, the founder and CEO of H2O.ai. “It’s like PageRank. Which are the most interesting features that people are using? Those are the concepts that we’re bringing [to the product], the idea of a recommendation engine for features.”

The feature store also includes bias detection capabilities, “just in case some variable about race or gender sneak into the alternative datasets, we create a score so that they de-bias their source data,” Ambati says. “So it’s making it safer.”

Don’t be surprised to see more co-development efforts such as this from H2O.ai and its customers, Ambati says. As companies begin to ramp up their enterprise AI efforts, they may find that H2O.ai has the right machine learning expertise to exploit their valuable data.

“We don’t have a lot of the data or the domain knowledge, and that’s where the customers come in,” Ambati tells Datanami. “They start building out customer applications, community-created apps, and some H2O-made apps as well, so then we can go to market together.”

Sri Ambati is the CEO and founder of H2O.ai

The vision and the goal, Ambati says, is to empower H2O.ai customers to become AI companies in their own right. “We’re beginning to now talk to several data companies and even unfractured companies because they do need to grow. Data is going to be how you write your code. AI has completely eaten software.”

Companies in specific industries that are competing with hyperscalers may be more inclined to partner with H2O.ai since they know that Ambati and company won’t compete with them.

“If you’re Walmart, you’re not doing business with AWS. Even if you’re in healthcare, you’re not fully sure the tech giants are not coming into your space,” he says. “All over, customers are embracing AI and starting to take on the tech giants, not just the hyperscalers, but even companies like Square and other are starting to come to the traditional banking space.”

You can learn more about the jointly developed feature store at www.h2o.ai/feature-store/.

Related Items:

Feature Stores Emerging as Must-Have Tech for Machine Learning

2021: The Year of the Feature Store

AWS Bolsters SageMaker with Data Prep, a Feature Store, and Pipelines