HPE Machine Learning (ML) Ops

2019: Announcement of the software container solution of HPE ML Ops

On September 10, 2019 the Hewlett Packard Enterprise (HPE) company announced the software container solution of HPE ML Ops created for support of complete lifecycle of model of the machine learning (ML) for local placement and cloud – public or hybrid. The solution represents the process similar to DevOps, for standardization of workflows of machine learning and acceleration of deployment of the systems of the artificial intelligence (AI) from several months to several days.

File:HPE ML Ops.jpg

According to the developer, the solution of HPE ML Ops expands possibilities of a software platform of BlueData EPIC using program containers with that providing to intelligent data analysis specialists access on demand to container environments for work with the distributed AI/MO and analytics.

As noted in HPE, in four years implementation of AI in corporations grew more than twice, and the organizations continue to invest considerable means and time in creation of models of machine and deep learning for a broad spectrum of use of AI, such as detection of fraudulent activity, the personalized medicine and predictive analytics of a consumer behavior. However the biggest call which technical specialists face known also as "the problem of the last mile" – is practical application of MO for successful deployment and management of the developed models and extraction from them commercial benefit. According to analysts of Gartner, by 2021, at least 50% of projects on MO will not be unrolled completely because of shortcomings of their practical application.

According to the statement of the developer, HPE ML Ops turns the technology initiatives connected with AI from experiments and pilot projects into production and business processes, covering all lifecycle of MO: from data preparation and creation of models before training, deployment, monitoring and their interaction.

"The models of machine learning only working bring to business commercial benefit. And thanks to HPE ML Ops we provide the solution of a corporate class allowing to implement complete lifecycle of machine learning for local placement and a hybrid cloud. We introduce the speed and flexibility of work of DevOps in MO, providing faster and cost-efficient use of AI at the enterprise", 'Kumar Sreekanti, the senior vice president and the technical director of division of Hybrid IT in HPE noted'

"From retail before banking, from production to health care and not only – practically all industries implement or investigate AI/MO for development of the innovation products and services for the sake of obtaining competitive advantage. While most the enterprises focus efforts on phases of creation and training of the projects in the field of AI/MO, they fight for practical application of complete lifecycle of MO – from the concept, to a pilot project, deployment in a productive system and to monitoring. HPE meets this lack, proposing the multiplatform solution for all lifecycle of MO on the basis of technology of containers intended for support of a number of operational requirements of MO, acceleration of obtaining results and achievement of good results in business", the vice president of the program of strategy of artificial intelligence in IDC marked out Ritu Jyoti,

"Our online games generate billions of data points (data points) every day. Applying the MO difficult models, our intelligent data analysis specialists use this information for the ordering analytics to improve play experience and to increase loyalty of our players. Using the software of HPE BlueData we containerize these analytical and MO of the environment to increase operational efficiency and to optimize our business", 'Alex Ryabov, the head of service of data processing in Wargaming noted'

According to the statement of HPE, using the solution of HPE ML Ops, the data processing specialists participating in creation and deployment of the MO models can feel possibilities of the most complex solution in the industry for operation and management of lifecycle of AI at the enterprise:

• Creation of model: previously prepared environments and the self-serviced sandboxes for testing of the MO tools and conducting records about data analysis;
• Training of model: scaled environments for training with secure access to data;
• Deployment of model: flexible and fast deployment with function of reproducibility;
• Monitoring of model: full overview of all lifecycle of the MO model;
• Interaction: the organization of workflows of continuous integration and delivery (CI/CD) using repositories for the code, models and projects;
• Security and control: the safe multi-user environment with integration into authentication mechanisms of the enterprise;
• Hybrid deployment: support of local placement and cloud – public or hybrid.

The solution of HPE ML Ops is compatible to a broad spectrum of the systems of machine and deep learning open source, including Keras, MXNet, PyTorch and TensorFlow and also to the MO commercial applications within software of a partner ecosystem, including on such platforms as Dataiku and H2O.ai, claim in HPE. As of September, 2019 the solution of HPE ML Ops is already available on a subscription.