Digital Catapult has today launched Machine Intelligence Garage, a program to develop artificial intelligence and machine learning start-ups in the UK.
Digital Catapult is a not-for-profit body that focuses on helping UK businesses to scale up. Its latest initiative, the Machine Intelligence Garage program, previewed in the UK government’s Industrial Strategy last week, aims to support the ambition of making the UK a global centre for artificial intelligence (AI) development. It’s an industry that politicians hope will add £232 billion to the UK economy by 2030.
According to Digital Catapult’s own survey of around 10 percent of the UK’s AI and machine learning (ML) start-ups, more than half are currently held back because they lack access to computation power.
Digital Catapult’s Machine Intelligence Garage programme, due to begin in January 2018, aims to tackle this issue by giving small and medium-sized enterprises (SMEs) access to cloud-based and physical computational power for AI. The organization is also promising to host a series of investor pitch days, meet-ups and showcases to create a knowledge-sharing environment among start-ups in this area.
According to Digital Catapult, deep-learning techniques, which have reached human-level performances, incur extortionately high computational costs – a single training run for a ML system can cost upwards of £10,000. This can be a serious barrier for UK innovators and researchers.
Clive Longbottom, analyst at IT advisory company Quocirca, said that when it comes to AI, ML and deep learning, it is very easy for huge amounts of resource to be required to reach a valid endpoint.
“If this is to be done on constrained equipment, then the results can take too long to come back. The costs of putting in place sufficient resource for real-time AI can be horrendous – therefore, it makes sense for a group to come together and enable a suitable environment for smaller start-up companies to access the resources they need,” he explained.
Longbottom suggested that, by sharing physical resources, risks can be shared too.
“For example, if there are 20 companies using a certain hardware system and a better one comes along, moving over to that new hardware is cheaper for everyone as a shared cost, rather than it all being down to a single player,” he said.
“Hopefully, the Machine Intelligence Garage will be a well-shared, multi-player environment with shared resources and shared risks across the board. Whether this will allow the UK to compete effectively against the massive R&D investments from the likes of Microsoft, IBM, AWS and Google in the US, or from low-cost or massively government-backed environments such as China, is something we will have to wait and see,” he added.
Digital Catapult is collaborating with a number of technology companies, high-performance computing facilities and academics for the three-year programme. These include Amazon Web Services (AWS), Google Cloud Platform, NVIDIA, Graphcore, STFC Hartree Centre, EPCC, Newcastle University, the Alan Turing Institute, Bart’s Health Trust and Capital Enterprise. The programme’s initial funders are InnovateUK and ERDF.
Companies developing products or services that use ML or AI can apply to take part in the initiative and applications will be assessed based on a number of criteria. These include strength of idea, technical implementation plan, availability of data, ethical use of data, and the immediacy of the need for computation power.
Applications for the programme will open every six weeks, with the first open call going live today. The first group of successful companies will join the programme on 23 January 2018.