Even small workloads get big results with machine learning on OCI

By Wei Han and Sasha Banks-Louie

Delivering on the promise of helping companies of all sizes make smarter decisions, machine-learning analytics continue to gain momentum.

While it’s still early days in 2022, companies large and small, and in nearly every industry worldwide, are quickly discovering the virtues of running ML workloads on Oracle Cloud Infrastructure.?

Oracle for Startups member, DigiFarm?has already helped some 14,000 farmers and agribusinesses across 30 countries determine their seeded field acre boundaries. Not only does this capability help growers forecast their expenses for seed, fertilizer, and insurance, it also helps them estimate their annual yields. In a recent episode of Built and Deployed, a technical video series for cloud architects, DigiFarm’s head of engineering, Rohit Shetty shared a reference architecture with Oracle senior cloud architect Mitesh Bhopale. ?Shetty’s architecture?illustrates how the Norway-based startup is optimizing crop production, using neural network models on Oracle Cloud. DigiFarm also uses Oracle bare metal GPU servers to train the deep neural network models on demand. The company uses Oracle serverless Functions that can scale dynamically to meet end-user needs. Not only does DigiFarm’s architecture help the company get better cost-per-performance in the cloud, but it also reduces operational overhead, and frees up the company’s engineers to focus on functional development.

In the CLOUDS Lab at the University of Melbourne, student researchers Mohammad Goudarzi and Qifan Deng helped move their FogBus2 research platform to OCI, allowing computer science researchers throughout Australia to capture, analyze, and make predictions off of Internet of Things (IoT) data. In a recent episode of Built and Deployed, which focuses on the University of Melbourne's?Machine learning models for IoT,?Goudarzi and Deng share their architecture with Oracle senior director of cloud engineering, Bill Wimsatt. The workflow starts with streaming computationally intensive and latency-sensitive IoT data to OCI, using Raspberry Pi and Jetson Nano edge devices. The ML workloads run on Ampere Arm compute cores, or Intel X86 processors, determined by how much compute power is required to process a given workload.

• Would you rather focus on application development? Then check out OCI’s?AI as a service, which includes language and anomaly detection services. These services have pre-trained ML models built-in and optimized for the job, so you don't need a data scientist or high-performance computing capabilities to leverage these AI functions.
• If you’re a cloud architect running ML workloads on OCI, we’d love to showcase your architecture on your own episode of Built and Deployed. Find out how to get featured here.

Check out these top 3 reference architectures from last week:?

Deploy a secure landing zone?

PayPal Essbase and analytics

Set up a Hub-and-Spoke network topology

Handpicked Reference Architectures from the OCI RA center

Use OCI Language for customer feedback analytics

By Gabriel Grigorie??????????

Use OCI Speech to transcribe natural language

By Wei Han

Deploy .NET applications on Oracle Cloud Infrastructure

By Maher Al Dabbas?????