Case Study: Allen Institute for Artificial Intelligence improves reproducible research with Google Cloud and Google Kubernetes Engine

Allen Institute for Artificial Intelligence runs 60,000 experiments with Google for Education

The Allen Institute for Artificial Intelligence (AI2) sought to overcome the cognitive overhead and infrastructure challenges that were slowing down its AI research. Their goal was to improve the reproducibility of complex experiments to accelerate the pace of scientific discovery. To address this, they partnered with Google for Education and leveraged Google Cloud, specifically Google Kubernetes Engine (GKE).

The solution was Beaker, a Kubernetes-based platform built on Google Cloud that automates the tracking of experiments, parameters, and results. By migrating to Google Kubernetes Engine, AI2 gained a stable, managed environment that freed its engineers to focus on delivering value. The results were significant: in 2018, researchers ran over 60,000 experiments totaling nearly 190,000 GPU compute hours, leading to at least twenty papers submitted to top journals. Google for Education provided the infrastructure that enabled this high-impact research.

Allen Institute for Artificial Intelligence