AI Insights
cancel
Showing results for 
Search instead for 
Did you mean: 

How to build AI solutions at scale in the enterprise

PankajGoyal

Using H2O Driverless AI with HPE servers and BlueData software can help businesses deploy AI solutions quickly and productively.

HP Ai-Blue Data-blog.jpgArtificial intelligence is already an indispensable technology for many businesses. Yet some enterprise organizations are struggling to keep pace with the competitive rush to harness AI and use it with their proprietary data to build predictive models. AI expertise is difficult to come by. And businesses often lack the experience needed to develop AI solutions from the ground up, and then deploy them at scale to deliver business impact.

Fortunately, help is at hand with the advent of machine learning solutions focused on addressing these challenges. New tools are hitting the market, and they promise to automatically create predictive models and help deploy them at scale. In this way, businesses can enjoy the benefits of AI while expending fewer resources.

H2O Driverless AI is one of the leading automated machine learning tools that helps provide such a solution. By tapping the processing power provided by multi-GPU servers from HPE, H2O Driverless AI enables companies to automate the creation of machine learning models and put them in production with ease—generating actionable predictions, customer insights, and business value.

Build AI models faster with H2O Driverless AI

To fully understand the benefits an automated machine learning platform can provide, it helps to look at a real-life use case. Suppose that you're a bank and you want to improve how you evaluate the credit risk scores of loan applicants. You can use artificial intelligence to do this, but learning how to build and test AI-based prediction models in-house takes a lot of time and investment.

Instead you could use H2O Driverless AI to build and deploy a robust machine learning model in a matter of hours—a process that would typically take weeks or months. By ingesting and digesting data related to customers' credit profiles and financial information, H2O Driverless AI quickly identifies the key features that predict a client's creditworthiness and creates new engineered features to enhance the accuracy of the model. These original and engineered features form a key part of the model to assess new applicants.

An efficient path to more accurate models

This is H2O Driverless AI at its simplest and most conceptual, however there are several other things the platform does to make deploying AI-based prediction models much less onerous, much more successful, and much more productive.

For example, its AutoViz feature lets data scientists assess the suitability of given data sets before models are being build and validated. AutoViz detects anomalies and outliers that could potentially affect the performance of models. It allows data scientists to get a clear understanding of their data sets before they start building models, or correct or change any data before advancing.

Once this preliminary stage has been completed, H2O Driverless AI is used to automatically generate hundreds of models. The system compares these models against one another to determine which are the most predictive.

In a medical context, H2O Driverless AI could compare models to identify which factors are the best at assessing a patient's risk for hospital readmission due to a chronic illness. Not only could the platform discover the most accurate model for, say, predicting readmission in urgent care, but it also provides machine learning interpretability that outlines the model's key features and variables in intuitive charts and graphs. This enables an organization's data scientists and other staff to interpret and comprehend any deployed model for themselves and for governance and regulatory compliance reasons.

Quickly deploy and scale AI environments with containerized environments

Using H2O Driverless AI in conjunction with HPE’s BlueData software also makes deploying large-scale AI-driven models easier.

The container-based BlueData software platform enables organizations to set up containerized environments where multiple different data science project teams can deploy machine learning models for their own specific use cases, running on shared infrastructure. This spares companies from the resource-intensive and time-consuming burden of installing the dedicated computing infrastructure for each specific use case. Instead, they can use the BlueData platform to adapt and allocate on-demand resources for each data science project, spinning up new containerized machine learning environments within minutes. They can also quickly and securely tap into enterprise data sources to train their models.

To use another banking example, the BlueData platform can enable financial institutions to use H2O Driverless AI in one division (e.g. retail banking) and allocate GPUs and other infrastructure resources for a specific machine learning project (e.g. to building a model to predict customer churn). Once that model has been built and trained, the institution is able to quickly release and re-allocate the GPUs and other resources to another group (e.g. the credit card division) so it can create a model for predicting transaction fraud. This can be repeated indefinitely. And the financial institution can reuse and repurpose the same shared infrastructure resources without having to implement dedicated infrastructure in every use case. The result is greater infrastructure utilization, higher return on investment, and reduced costs.

Scale up with HPE servers

Together, H2O Driverless AI and BlueData can accelerate the deployment of AI throughout an enterprise organization and make it more cost effective. Using the HPE ProLiant DL380 with support of up to 3 NVIDIA® V100 32GB GPUs, or the HPE Apollo 6500 servers with support of up to 8 NVIDIA® V100 32GB GPUs, provides additional efficiency and power. This is how H2O Driverless AI is optimized to exploit GPU acceleration to speed up the automated machine learning process.

Because H2O Driverless AI includes support for such GPU-accelerated algorithms as TensorFlow, XGBoost, and LightGBM, the ProLiant DL380, and the Apollo 6500 servers opens up myriad possibilities for developing AI-powered models out of proprietary data sets.

The artificial intelligence software market is expected to be valued at around $118 billion by 2025. Given the insights provided by AI solutions like H2O Driverless AI, HPE’s BlueData software, and HPE servers, it's not difficult to see why.

These tools make it easier for businesses to build and deploy their own machine learning models. Thanks to the power of HPE servers, these models can be used at scale. Such AI solutions prevent many of the headaches that come with building AI models and programs manually. Ultimately, they can help almost any kind of organization obtain concrete results when deploying AI.


Pankaj Goyal
Vice President, HPE AI Business
Hewlett Packard Enterprise

twitter.com/pango
linkedin.com/in/goyalpankaj
hpe.com/ai

0 Kudos
About the Author

PankajGoyal

Pankaj is building HPE’s Artificial Intelligence business. He is excited by the potential of AI to improve our lives, and believes HPE has a huge role to play. In his past life, he has been a computer science engineer, an entrepreneur, and a strategy consultant. Reach out to him to discuss everything AI @HPE.

Labels