Machine learning & AI

We’re working at the frontier of this vital, transformative technology, developing systems that leverage deep learning techniques to achieve unprecedented performance in a huge range of applications.

From detecting anomalies and attacks in networks and learning to control legacy systems to high-performance audio and image processing and interpretation, optimising infrastructure deployment and natural communication with humans, the possibilities are endless.

The Digital Greenhouse is a purpose-built facility designed for discovering, developing and testing machine learning approaches in a fast, secure environment.

Inspired by botanical sciences, our team ‘grows’ hundreds of strains of models to understand where the richest pickings are to be found – beyond the confines of ordinary software development methodologies. New algorithmic hopefuls are assessed against long-standing commercial and industrial challenges, such as optimising the deployment of cellular infrastructure or detecting anomalies on a manufacturing line.

The Digital Greenhouse runs on high-performance computing based around NVIDIA’s DGX-1 Deep Learning Supercomputer and other GPU- and FPGA-accelerated servers, providing petaflop-scale compute on-site.

This links to petabyte-scale local storage, project-specific clouds and our continuous integration systems. When our organically grown machine learning is ready, we can export models easily to customers’ own compute facilities or the cloud.

Our machine learning expertise spans older, established techniques such as Bayesian inference and support vector machines, through to the latest advances in deep learning, generative networks, natural language processing and unsupervised learning.

We can design, develop and deploy sophisticated bespoke machine learning systems from the ground up or work with third-party services and platforms. Particular specialities include real-time/low-latency processing at the network edge, lower power silicon design for machine learning and tools that allow deep learning to be trained effectively with much smaller datasets than is typically believed possible.