It’s no surprise that companies across industries are turning to AI and machine learning. The worldwide AI market is forecast to reach $327.5 billion in 2021 and exceed $500 billion by 2024, according to IDC.
And yet, while many businesses know the importance of using AI to stay competitive, AI adoption isn’t always easy – it’s a comprehensive process, from data ingestion to model monitoring, and each step poses its own challenges.
In Oracle Developer Live: AI and ML for Your Enterprise, Oracle experts shared their insights on how to use AI and machine learning for any business. Here are the key takeaways.
Most companies are not in the core business of building AI models, said Elad Ziklik, vice president of product management for AI services and data science. Ziklik compared AI to Legos – many companies want a Lego kit, a complete set of all the tools and instructions they need for a specific use case, instead of a disparate set of building blocks with no clear instructions.
As a result, AI will trend toward domain-specific models for industries like financial services and manufacturing. However, these models require a lot of data and domain expertise – both of which Oracle has.
“There’s only one software company that has both a general-purpose cloud platform and is a world leader in SaaS apps and domain-specific apps, and that’s Oracle,” he said.
With that, he introduced the upcoming Oracle Cloud Infrastructure (OCI) AI Services, which allow companies to apply pretrained AI models to their business without needing in-house AI or ML experts.
Learn more about AI Services in Ziklik's Developer Live keynote with Manisha Gupta, senior director of product management for analytics apps for HCM. To gain early access to these services, email email@example.com.
Many speakers at Developer Live agreed that it’s important to define the business challenge first, so that your team can identify the right data and models for your use case.
“It beings with the definition of the problem, and then having the right data,” said Suhas Uliyar, vice president of product management.
“Start thinking about what problems you may want to solve in the future, and how you can start to prepare for them now, collecting, categorizing data,” said Ian Wilson, director of engineering for IntelligentSuite.
Watch this panel to learn from Oracle leaders on how to apply AI toward your business goals.
The machine learning lifecycle requires more than just building models. Over 50% of the time on AI projects is spent integrating and managing data vs. executing data science tasks, according to an IDC and Oracle report.
“A lot of focus today is on algorithms, and the different set of models that are available, and the services available, but I actually believe that data is a new programming language – that is, you begin to control the AI model by controlling its training data,” said Uliyar.
First, it’s important to know what data you have in order to use it best. Abhiram Gujjewar, director of product management, spoke about OCI Data Catalog, a metadata management solution that makes it easier to find valuable trusted data in the enterprise.
Data science teams also need to prepare and transform data until it’s ready for modeling. Carter Shanklin, senior director of product management, and Julien Testut, senior principal product manager, highlighted two solutions for more easily integrating and preparing data for data science – OCI Data Integration, a fully managed serverless ETL service, and OCI Data Flow, which helps customers deliver Apache Spark-based applications faster.
“While a significant part of the machine learning process can be automated, there are limits we face today,” said Mark Hornick, senior director of data science and machine learning product management.
For example, AutoML eliminates repetitive tasks in model building, so that teams can increase productivity and reduce compute costs. However, business problem definition and data understanding require a human perspective; people bring domain-specific knowledge and define success criteria to these steps, Hornick noted.
“This AutoML engine does automated algorithm selection, it does data and feature selection for you, and automated hyperparameter tuning," said Elena Sunshine, director of product management. “So you can essentially take a dataset and spit out a really good candidate for a model for that data within a very short amount of time with very low effort.”
Essentially, AutoML's role is to enhance, not replace, data scientists' work.
See how OCI Data Science incorporates AutoML and other features to empower data scientists at every stage of the machine learning lifecycle.
Once you’ve built the model, the process isn’t finished – you also need to deploy that model, and monitor it while it’s in production. Marcos Aranciba, product manager for data science and big data, defines machine learning models as math functions, and “model deployment is translating that math formula into a result.”
He cites key challenges for machine learning deployment, including having a model developed in a different environment from where it’s deployed, and needing to integrate models into end user applications. You can learn more about challenges in machine learning deployment, and how Oracle Machine Learning addresses these challenges, in this presentation.
Finally, AI is not limited to data scientists. Anyone who consumes AI, from developers to business users, can get involved.
Hornick recommends that developers learn about machine learning concepts and algorithms. “This will better prepare you to collaborate with the data science teams in your organization,” he said.
Ziklik also noted in his keynote that, by providing pretrained AI models, the OCI AI Services aim to "empower data science teams to effectively collaborate with developers, data engineers, and operators, and deliver AI-powered solutions."
To view all the presentations from Developer Live, catch the replays of Developer Live: AI and ML for Your Enterprise.
Want to try the services featured in Developer Live? Sign up for a free Oracle Cloud trial.