What is Deep Learning?
Unsurpassed capability
Deep Learning is a subset of Artificial Intelligence, processing complex multivariate data sets through the use of deep multi layer networks, such as neural networks.
Applying the most appropriate algorithms, topologies, and architectures, it provides exceptionally powerful advantages and solutions for a range of Insurance & Reinsurance business opportunities.
Its capability is unsurpassed, extracting knowledge from all manner of data, including unstructured data recorded in text, image, and audio formats from traditional sources such as documents or new ones such as Internet of Things sensors.
This capability enables accurate predictions and hidden patterns, insights and causal inferences to be revealed. It can connect the dots hidden in data faster than any person.
Previously, such capability was only possible, though in a much more limited way, using related Artificial Intelligence – Machine Learning or Mathematical – Statistical Learning methods. Deep Learning’s methods are equivalent to those of these alternative approaches but use different algorithms, techniques, and technology and achieve better outcomes.
Powerful versatile methods
Deep Learning is significantly more powerful and versatile than alternative Machine Learning and Statistical Learning methods.
It can analyse much larger and more complex multi dimensional and rich datasets. The other learning methods are more constrained to smaller, lower dimension, often linear, models.
Deep Learning is faster and more accurate, particularly when consuming large datasets, scaling linearly rather than plateauing.
While Machine Learning and Statistical Learning methods have been used for many years; they have tended to be used in more specialised and narrower areas by personnel such as actuaries and quantitative analysts. Deep Learning can be used at all levels throughout an organisation, it should be available to all.
Advanced technology
Deep Learning uses the latest high performance parallel graphical processing unit (GPU) technologies from companies such as NVIDIA and IBM.
Performance is continually improving. A recently released hardware server with 16 interconnected GPUs can deliver 2 petaFLOPS of computational power as well as being able to consume even bigger data models.
Our approach is to exploit the optimal and most appropriate high performance computing architecture for Deep Learning. Currently GPU technology, optimised for Deep Learning, outstrips that of traditional CPUs. For example, the NVIDIA DGX station GPU desktop machine is faster by a factor of x47 than a CPU server.
