putilov_denis - stock.adobe.com
Hewlett Packard Enterprise bolstered its growing portfolio of artificial intelligence technology Monday with the acquisition of a San Francisco-based startup that specializes in software to train AI models.
HPE CEO Antonio Neri said the purchase of Determined AI would enable customers to deploy "ready-to-run" training models for AI in systems at massive scale with the high-performance computing (HPC) systems of Cray, a 2019 HPE acquisition.
"HPC is a space where we dominate today," Neri said, "and we think this will accelerate that position in the market."
Two of the greatest challenges Neri said customers face with AI and deep learning are a lack of expertise and the extensive data preparation that happens before they can architect and run algorithms. Using Determined AI's machine learning technology on top of Cray's programming environment for AI simulation and modeling could help ease the work and let customers focus on getting more business value from their data, Neri said.
Founded in 2017, Determined AI launched an open source deep-learning training platform in 2020 and also sold a full-featured enterprise product for teams to train models at higher scale. The Determined AI acquisition is the latest in a string of HPE AI-related purchases that includes BlueData, which closed in 2019, and MapR, also in 2019.
Complement to HPE Ezmeral offerings
Chirag Dekate, a vice president and analyst at Gartner, said Determined AI would complement the BlueData and MapR technologies that comprise the core of HPE's Ezmeral Machine Learning Ops (ML Ops) and Ezmeral Data Fabric offerings, respectively, and give HPE a strong portfolio of products to facilitate enterprise AI strategies in line with its hybrid, multi-cloud vision.
Chirag DekateVice president and analyst, Gartner
"We are seeing enterprises accelerate their operationalization of AI and build AI orchestration and automation platforms," Dekate said. "And what you're starting to see emerge within the HPE strategy is an end-to-end ecosystem that aligns quite well with an AI orchestration and automation platform."
Dekate said the Determined AI technology decouples model development from the underlying infrastructure and lets enterprises use any infrastructure they want, whether on premises, in the cloud or a combination of the two. Using the Determined AI software, teams of data scientists can work side by side using disparate frameworks such as TensorFlow, PyTorch or Keras to develop AI applications across a shared infrastructure pool, he said.
"When enterprises are starting out with an AI initiative, they tend to have small teams dedicated toward one or two projects," Dekate said. "But as enterprises mature, they are working on several dozen projects at the same time, in many cases, and they have a team of data scientists working on their own projects in a collaborative space."
Key Determined AI capabilities
Dekate said Determined AI stood out among 70 or 80 startups in the AI market with its well-known engineering team and focus on enabling enterprises to employ a hybrid, multi-cloud environment to accelerate deep learning model training, validation and deployment. He cited Determined AI software capabilities such as cluster sharing and resource management; automated hyperparameter search and optimization to help developers identify the right models faster and more efficiently; distributed training to use diverse resources; and experiment logging and tracking through a central dashboard.
"As with any acquisition, we will be watching it very closely to see how successfully they integrate the product and how they expose the unified capabilities to end users," Dekate said. "But, frankly speaking, Determined AI is quite complementary to HPE Ezmeral ML Ops and, I think, you will see them work quite well with one another."
Kashyap Kompella, CEO and chief analyst at RPA2AI Research, said Determined AI is a "good tuck-in acquisition" for HPE that fits well with its HPC and cloud strategies. He said HPE is a leader in the HPC segment, where customers tend to prefer private clouds and HPE has a good presence with its GreenLake as-a-service offerings.
"While much of academic research on AI is focused on improving AI methods and models, the Determined AI team is known for their interesting research on the practical aspects of AI, such as optimizing and deploying deep-learning models into production," Kompella said.
He added, as the AI models used in real-world applications get bigger, they require huge computing infrastructure. Costs can run to millions of dollars as the engineering aspects of training and deploying AI models grow more complex, according to Kompella.
"With Determined AI technology," he said, "HPE must be looking to cut down those bills for their customers."
Carol Sliwa is a TechTarget senior writer covering storage arrays and drives, flash and memory technologies, and enterprise architecture.