Analysis: existing technologies will be unable to meet future internet needs, but developments in AI could help connect rural areas and lower costs
Artificial intelligence (AI) was centre stage at the World Economic Forum in China. At least 20 of the 56 companies selected for the organisation's Technology Pioneers programme are using AI in some way with applications ranging from autonomous vehicles to advertising technology.
A branch of AI, machine learning, is dedicated to the ability of a machine to learn something without having to be programmed for that specific thing. It enables computers to improve their performance automatically over time by being fed data and information in the form of observations and real-world interactions – like a toddler learning about the world around them. Answering whether the animal in a photo is a cat or a dog, spotting obstacles in front of a self-driving car, spam mail detection, and speech recognition of a YouTube video to generate captions are just a few examples out of a plethora of predictive machine learning models.
So, how can machine learning help in transforming our communications networks?
From RTÉ Radio 1, Drivetime's Philip Boucher-Hayes reports on the cost governments around the world have paid for high-speed broadband
To appreciate its usefulness for this fast-growing application, we need to understand how the Internet is operating nowadays. Well, almost all the data carried by the Internet is transmitted over fibre-optic cables. These fibres, made of a high-tech glass, allow data to travel over long distances at nearly the speed of light (300,000 km per second) providing us with the high-speed broadband Internet we now take for granted.
But the popularity of modern services such as Netflix and YouTube results in increased data traffic and this is placing these cables under huge pressure to meet our demands. This is known as the 'capacity crunch’ and existing fibre-optic technologies will be unable to meet our future Internet needs and expectations. To be more specific, the main limitation on Internet speeds is that data signals propagating over the optical fibre are distorted. The longer the optical fibre, the more distortions we have due to the glass of the fibre.
Hitherto, scientists have partially solved the problem using digital, software coding that enables signal distortion correction. However, the trouble that scientists are facing is that a significant portion of these distortions are completely random and, as expected, very difficult to address.
Among the most valuable machine learning models to address this problem is the artificial neural network (ANN) which attempts to simulate the complexity and power of the human brain. The great benefit of a neural network is that we don’t have to program it to learn: it learns all by itself. It just needs to be exposed to data, like the brain.
From RTÉ Radio 1 Today with Miriam, almost half of Irish CEO's believe AI will ultimately replace more jobs than it will create
So, how ANN could help our Internet networks?
ANN provides an advanced statistical approach with the powerful ability to deal with these troublesome random signal alterations by 'training’ signals. This a key feature in ANN, which is dedicated for making predictions by feeding a lot of data into the model and permitting ANN itself to learn more about the processed information. Whilst ANN also functions as a digital fix, programmed in the electronic telecommunication modems, using this ‘training’ process to gather observed historical data about the fiber-optic network it "learns" about network performance impairments, building a probabilistic model.
Using this model as a reference, its artificial neurons are then responsible for deciding whether and how distorted data signals can be repaired, essentially operating as ‘digital filters’. Another benefit of ANN is that it is of much lower complexity than state-of-the-art techniques, therefore offering an energy and cost efficient solution.
Today's state-of-the-art is simply not powerful and effective enough
Indeed, it is now possible to develop modems which incorporate ANN technology to provide consistent high-speed broadband connectivity. At CONNECT, the Science Foundation Ireland research centre for future networks in Dublin City University, we are working towards the commercialisation of this idea. Our work explores the use of ANN and other machine learning techniques in optical fibre telecommunications to tackle the random distortions in the network and substantially improve the signal transmission quality.
The impact of this work and further developments in machine learning could be significant. It could help with the connection of rural areas by facilitating the use of longer optical fibres without any additional hardware, and with an associated cost saving. This could also be of use in developing countries which, though not yet facing the capacity crunch, are increasingly networked and need affordable technologies to enable participation in the global economy.
From RTÉ 2fm Dave Fanning, Adrian Weckler, technology editor of the Irish Independent, on how 5G works and why it's needed
Machine learning is also expected to be the key to sustaining a low-latency telecommunication network. The ability to process a huge volume of data with minimal delay, known as ‘low latency’, is at the heart of 5G - the next generation of mobile communications networks. Machine Learning will be key to sustaining this and enabling the two-way communications which will be essential for smart city applications such as driverless cars.
Though machine learning has been implemented already in certain areas, such as in medical diagnosis, robotics, and marketing, its application to the world of telecommunications will be more challenging because of the time-critical nature of typical communications: machine learning has only milliseconds to respond when a problem is encountered, for instance, when a stock-trading financial transaction is taking place.
From RTÉ Radio 1's Today with Sean O'Rourke, Brian O'Connell visits the counties with the slowest broadband speeds
Therefore, more research is required regarding practical machine learning algorithms that can analyse huge and complex data while delivering accurate results at a faster rate. Machine learning methods have the potential to replace traditional communications and signal processing systems in the future: today’s state-of-the-art is simply not powerful and effective enough.
Machine learning looks set to shift from a position of prominence to one of dominance in this space.
CONNECT will host a summer school on Machine Learning for Communication Systems and Networks at Trinity College Dublin between September 2nd and 4th
The views expressed here are those of the author and do not represent or reflect the views of RTÉ