On the business front, enterprises that could adapt quickly to a changing business environment are in the best position to ensure business continuity and long term profitability. Digital businesses and enterprises that are further along with their digital transformation journeys are better equipped to respond to this rapidly changing environment. At the core of this transformation is the ability to leverage data to deliver actionable insights and predictions using machine learning and artificial intelligence.
Prior to the pandemic, enterprises were already adopting machine learning (ML) and artificial intelligence technologies at a rapid pace; this adoption has been further accelerated by recent global economic and social changes. But experimenting with ML is the easy part. The harder part is integrating ML models into business applications and processes to scale ML across the enterprise. In a recent survey conducted by Forrester Research, 97% of decision-makers reported that not just ML but MLOps—having mature processes to deploy and operationalize machine learning at speed and scale—is essential to success.[1]
ML offers solutions, but…
ML has proven valuable in making sense of vast quantities and varieties of data. But as promising as machine learning is in solving the critical and urgent needs of enterprises—data science teams continue to face challenges when it comes to effectively deploying AI and ML models. In fact, only 6% of those surveyed in the Forrester study reported they have mature MLOps processes[2]. An overwhelming majority cite a lack of tools and technology to support ML operationalization as one of the top reasons for their lack of operational ML workflows.
Data science organizations face three common challenges as they operationalize and scale their machine learning models:
Access to data: Machine learning models are trained on vast amounts of data from multiple different sources. Data scientists explore datasets and join data from various sources to find the right set of data that best represents the world they are trying to predict.
Ensuring data scientists have access to the most current data with security, privacy, and access control guardrails in place is an essential requirement for ML in the enterprise. Data scientists often copy data into local environments and expose enterprises to risk on multiple fronts—the risk of regulatory non-compliance, the risk of incorrect predictions, and the risk of a data breach. Enterprises need to plan for the growth in data and deploy the right infrastructure (software, hardware, and/or cloud services) to ingest, store, and curate that data. They also need to ensure this data is available to their data science teams.
Access to flexible and scalable computing environments: The field of ML is rapidly evolving, and different business problems require different tools. The ML model training process also requires access to scalable compute environments with specialized infrastructure—CPUs, GPUs, and other hardware accelerators. At the same time, data scientists need to experiment with different ML tools and frameworks and pick the one best suited to their business problem.
IT teams that support data scientists face challenges with the traditional processes of provisioning clusters on bare-metal servers or VMs. The result is the perception of IT as a bottleneck. In reality, IT just doesn't have the resources to keep up with the various tools and technologies required by data scientists, while also ensuring security and access controls are enforced around data and compute resources.
Standardized collaboration processes: Due to the experimental nature of data science, data scientists typically develop applications on their laptops or in local environments. This process creates silos that limit collaboration between the data science, data engineering, and software engineering teams and between different data science teams. One of the impacts of this disconnect is that it could take several months for a model to be deployed to production, thereby severely limiting its effectiveness.
A lack of standardization around collaboration also means that enterprises rely on tribal knowledge to determine the datasets, transformations, and code necessary to retrain and redeploy the models when it comes to retraining the model. Data science processes need to provide the flexibility to experiment while also ensuring collaboration and sharing of learnings amongst team members. This flexibility can help avoid siloed and isolated development environments that inhibit the sharing of best practices across teams.
HPE addresses the challenges with operationalizing ML
At HPE, we understand these challenges. We have helped leading enterprises in every industry scale up their ML and data science initiatives, and we know what it takes to build operational ML processes and workflows. We've integrated these learnings into our container-based MLOps software platform that we now offer under the newly announced HPE Ezmeral software brand: HPE Ezmeral ML Ops.
HPE Ezmeral ML Ops is an enterprise-grade solution that brings speed and agility to the ML lifecycle, while also standardizing processes and enabling seamless collaboration between data scientists, data engineers, software developers, and IT teams. By automating tasks related to cluster provisioning and management, HPE Ezmeral ML Ops allows researchers and scientists to focus all their effort on building models and accelerate the delivery of ML-based solutions that improve business outcomes.
HPE Ezmeral ML Ops is a solution built on the HPE Ezmeral Container Platform. Leveraging 100% open-source Kubernetes, HPE Ezmeral ML Ops works with the entire Kubernetes ecosystem of tools and technologies for ML and data science–as well as other open source and commercial products in the AI / ML ecosystem–providing the flexibility to deploy containerized environments for whatever tools your data science teams prefer.
HPE Ezmeral ML Ops is available as standalone licensed software to run on any infrastructure (with HPE or non-HPE hardware), on any public cloud, or at the edge. And today we announced that our MLOps solution will be available as a pay-per-use cloud service deployed on-premises through HPE GreenLake.
We also provide HPE Pointnext Services to help our customers accelerate their AI / ML deployments at every step, with advisory and consulting services as well as deployment and support services to ensure customer success.
Find Out More at HPE Discover 2020
HPE Discover 2020 is a free, ongoing, virtual event that started on June 23rd, and offered a rich collection of live and on-demand sessions and demos related to HPE Ezmeral. Now available on-demand, visit the links below to view on-demand sessions and demos related to HPE Ezmeral ML Ops, such as:
- Software Spotlight LIVE BROADCAST: HPE Ezmeral: Software to Fuel Data-Driven Transformation (Kumar Sreekanti, Robert Christiansen, Ted Dunning, Ellen Friedman, Sunil James, et al.)
- Bridging the Chasm Between Model Development and Production with MLOps (Nanda Vijaydev)
- MLOps: Industrialize ML Development to Deliver on the Promised Value of AI (Matt Maccaux)
- Accelerating Business Value Through MLOps: An Applied Perspective (Gaurav Kaul)
- Accelerating Innovation with a Broad ISV Partner Ecosystem (Ron Kafka)
- kubectl Get *Ops (Matt Maccaux)
- HPE Ezmeral Container Platform as a Service: Optimized for Splunk (Matt Maccaux)
- Take an Interactive Tour of the HPE Ezmeral Container Platform (Don Wake, Terry Chiang)
- Delivering an as-a-Service Experience for GPU-Accelerated Workloads (Tom Phelan)
- New IP from HPE: Containers and MLOps as a Service with HPE GreenLake (Nanda Vijaydev)
Other sessions; not yet available on-demand:
- W686: Hack Shack (LIVE): Deploying End-to-End Machine Learning Workflows (multiple sessions available)
- T492: Hack Shack: Accelerate Innovation with DevOps for Machine Learning
Interested in other sessions? Check out the full line-up of sessions from HPE Discover Virtual Experience 2020.
[1, 2]Forrester, Operationalize Machine Learning, June 2020
Matheen Raza
Hewlett Packard Enterprise
twitter.com/HPE_Ezmeral
linkedin.com/showcase/hpe-ezmeral
hpe.com/ezmeral, hpe.com/mlops
Matheen_Raza
Matheen Raza is a Senior Product Marketing Manager at Hewlett Packard Enterprise (HPE) in the Enterprise Software Business Unit. A technology enthusiast, he is extremely passionate about cutting edge technology such as containers, AI, machine learning, and deep learning to improve business outcomes in enterprises.
