Machine Learning as a Service Cloud for REG.COM

REG.COM is one of the biggest ICANN-accredited domain registrars and hosting providers in Europe. When REG.COM approached Aligned Research Group, REG.COM had multiple businesses, such as domain registration in 220+ domain zones, hosting, e-mail, VPS/VDS services, website builder, to name a few. But what REG.COM, although successful in their area of expertise, lacked experience with GPGPUs, and explored the possibility to add Machine Learning as a Service (MLaaS) to their portfolio of products.

Aligned Research had previously optimized a number of algorithms for GPGPU, including contributing to open source R (we re-wrote robust linear regression using NVIDIA GPGPU with a significant speed-up). But what differentiated us from other companies good at GPGPU was our DevOps expertise, which we combined with our GPGPU knowledge to suggest creating a proof of concept of an analytical platform suitable for both scientific and analytical customer loads, “Machine Learning as a Service”.

The architecture included both NVIDIA-supplied and custom Docker containers on top of the Kubernetes cluster with fine-tuned volume management, networking, and a modified runtime together with a set of basic monitoring tools.

To show the value of the platform, we developed a variety of template notebooks for Jupyter platform. It included a list of pre-trained models (FaceNet, AlexNet, and Yolo) and multiple initial datasets (ImageNet, ОpenImage, and Stanford Sentiment TreeBank) to shorten the learning curve of the users of this MLaaS platform.

The prototype was a resound success and exceeded REG.COM expectations. The company, upon testing users’ feedback, decided to use this service to partner with several universities to increase awareness of their cloud among students and faculty.

This project is one of many when we helped established companies to speed up innovation cycle, as it is often difficult for a corporation to make a decision of a large scale implementation before quickly creating a proof of concept or a prototype. When unusual for a core expertise skill is required, going with the right provider of “innovation as a service” may speed up the cycle from innovation to customers and make it less risky, compared to when the company tries to hire for this specialized skill on the labor market. When the project does not continue after the proof of concept, direct hire of the rare skill can backfire. If the prototype is successful, a longer-term staffing plan can be executed with a clear strategy. Employing “fail fast” strategy with Aligned Research Group, corporations can emulate agility of a startup within their established ecosystem and maintain high morale of their core talent.