BackSearchSearch by voice
 

University of Oxford

Oracle Customer Success

University of Oxford researchers predict career pay using Oracle Cloud

Summary

An app created by the University of Oxford’s Dr. Matthias Qian and powered by OCI Data Science and Oracle for Research, helps job seekers by predicting salaries, regardless of location. 

Business challenges

Dr. Qian introduces deep learning to economics research, aiming to help job seekers make informed decisions based on salary differences for the same job, no matter the location.

To train their models, Oxford researchers needed powerful compute and storage that they could easily scale. They also sought a technology partner who understood the business challenge and could share proven practices and services. Finally, they didn’t want to manage their own GPU servers.

They turned to Oracle for Research, which supported Qian and his team with a one-year Oracle for Research grant, providing access to Oracle Cloud and technical advising and collaboration.

We decided to use Oracle Cloud Infrastructure Data Science Service because it allows us to leapfrog the management of GPU servers, including the installation of the CUDA drivers, and to scale our compute resources on demand.

Dr. Matthias Qian

Departmental Lecturer, Department of Economics, University of Oxford

Why University of Oxford Chose Oracle

Oxford researchers chose Oracle Cloud Infrastructure Data Science because they could easily scale compute resources without needing to manage GPU servers. They liked the platform’s simplicity and collaborative, project-driven environment. They benefited from its unique features for data profiling, model development, and model explanation, as well as the flexibility of using various algorithms and frameworks—all things they did not find in other cloud providers.

Results

Working with US-based Burning Glass Technologies, University of Oxford researchers, led by Dr. Matthias Qian, collected a million job ads across many industries and cities in order to predict the future of work.

Using Oracle Cloud Infrastructure Data Science, they trained neural network algorithms to create a salary prediction model based on this data, as well as predict the adoption of artificial intelligence across professions. They then developed a web-based interface to allow users to interact with the model and explore different geographical job markets. Ultimately, their research aims to create a predictive deep-learning model that can give job seekers and policy makers a glimpse into the future of work and the workforce.

Predictive neural net algorithms require massive, high-speed compute—something that most researchers don’t have ready access to, especially on-demand. The ability to test and train a model while keeping data private is also critical. With easy and secure access to HPC and GPU shapes, Oracle Cloud Infrastructure is ideally suited for this type of research.

In addition, researchers using OCI Data Science notebooks can integrate new and popular data science concepts such as long short-term memory and transformer-based neural networks. Working collaboratively, with as little or as much compute control as they like, they can tune these to ensure the model is behaving as expected.

“One of my researchers uses Python scripts in the OCI command line shell to start GPU servers, whereas I often prefer to use the OCI Data Science platform, because it gives you really quick access to computing resources,” says Quian. “I usually don't have time to finish an analysis I started, so I pass it on to other team members who can take a deep dive. As a data science platform, this flexibility is perfect for our needs.”

Partners

Oracle for Research is a global community that is working to address complex problems and drive meaningful change in the world. The program provides scientists, researchers, and university innovators with high-value, cost-effective cloud technologies, participation in Oracle research user community, and access to Oracle’s technical support network. Through the program’s free cloud credits, users can leverage Oracle’s proven technology and infrastructure while keeping research-developed IP private and secure.

I value Oracle resources for its scalability. One of our researchers needed a service with more than 200 gigabytes of memory in order to process mobile colocation data. It took me just 10 minutes to provide this for him, and I didn't need to shop around for a new fleet of computers.

Dr. Matthias Qian

Departmental Lecturer, Department of Economics, University of Oxford

Published: February 16, 2021