Computer Science News
Cogent Labs and University College London Partner in Long-term Joint Research Initiative
Cogent Labs and UCL will pursue a unique joint research approach where AI research will be long term and ongoing, fully collaborative with researchers from both institutions working together, with all results being made public and shared with the wider AI community and society at large.
UCL researchers bring extensive AI expertise and established relationships with the international academic community while CL, in addition to its research team, will contribute with its strong machine learning team lead by David Cournapeau, the original author of the globally used machine learning library SciKit Learn. Cogent Labs will provide unique datasets accumulated through its ongoing business with Japanese clients allowing the researchers to apply their work to real world business problems.
UCL and Cogent Labs joint research will be conducted via the new UCL Centre for Artificial intelligence, led by its Director, Dr. David Barber. The Centre will be formally launched in early 2019, and will encompass ALL of UCL’s Computer Science AI activities in one space for the first time. Comprising over 150 AI researchers with a shared interest in addressing fundamental AI challenges, the UCL Centre for Artificial Intelligence's vision is to create new technologies and advise on their application and impact in science, industry and society. The new Centre for AI provides an exciting and novel opportunity to engage in a wide range of projects that span across traditional AI boundaries. This collaboration with Cogent Labs will foster knowledge exchange across the breadth of expertise within UCL's Centre for AI.
The human brain is capable of solving a wide range of very different types of tasks. With current AI systems on the other hand, it is necessary to create and train a separate neural networks for each type of specialist task. This presents a major obstacle in the advancement and adoption of AI.
One striking feature of human reasoning capabilities is how we can effortlessly learn to decompose sophisticated varying problems into sequences of simpler tasks that can be solved readily. Neural networks currently cannot do that, they are monolithic and have to be trained for a specialised task.
The ambitious goal of our collaboration with UCL is to understand how to design neural networks in such a way that modular solutions to complex problems naturally emerge. The first in a series of results from the joint research effort is expected to be announced in 2019.
Accelerating the move towards more general AI
The development of this technology could make it easier to re-purpose a single neural architecture to solve different problems. There are many tasks for example which require perception but call for different actions. An autonomous vehicle needs to detect pedestrians, but the next action is to navigate and avoid them. Security cameras also detect people walking by, but would be expected to follow with very different actions. The modular ML approach is a research area that would allow similar neural architectures to be used in both these cases, but taking different actions depending on the context. Modularity will greatly increase generality of AI systems and make them considerably more powerful, easier to train and deploy.
Professor David Barber, Director of the UCL Centre for Artificial Intelligence, said: "The partnership with Cogent is hugely exciting. Whilst AI has made substantial progress, current machines still have a limited understanding of us and the world around them. Improving the ability of AI machines to be useful for us requires tackling the bigger issues, such as how to allow machines access to knowledge that we have and how to train machines based on only a limited amount of interaction. Our partnership with Cogent will support our efforts to address such longer term challenges and we're honored that Cogent have chosen UCL based on our research excellence in AI. As AI begins to play an increasing role, it's important that we develop AI openly for the benefit of science and society, shaping a brighter future for all. Cogent is an ideal partner with whom to share that vision."
Griff Jones, Head of Section and First Secretary for Science and Innovation at the Science and Technology Department of the British Embassy, added: "The British Embassy Tokyo is delighted that Cogent Labs and UCL have decided to form a research partnership on artificial intelligence. The UK and Japan have a long history of partnering in research that delivers significant benefits for businesses, researchers and both our countries."