The Importance Of Machine Learning For The Future Of AND Digital

27 July 2017 | Maher Atashfaraz | About a 4 minute read
Tags: Analyst, artificial intelligence, associate, concept, data, developer, Learning, machine learning, tech, technology, virtual reality

Over the past few years machine learning has been one of the largest focus in technology, gaining interest from global tech companies such as Google and Facebook. But what is machine learning and why it is such a big deal?


We can think of machine learning as a learning algorithm, which has the ability to learn from large quantities of data and produces accurate results from this. Regardless of the domain the underlying algorithm stays the same; the size of your dataset and the quality of data decides how well your AI (Artificial Intelligence) performs. The bigger and higher quality datasets tend to create a more robust, intelligent AI. As an example I will briefly explain my final year dissertation project which revolved around using machine learning – or more specifically deep learning in image recognition. The project was to design a game of rock, paper, scissors where the user could play against virtual agent visually with a camera. You can find the code and showcase of it in the links here and here.


In order to create my virtual agent, I wanted to create an AI system that was able to recognise hand gestures. The concept was given an image or a set of camera images of hand gestures, for it to recognise the player’s move. To train my AI, I had to create a dataset for it to learn from – in this case the relevant hand gestures being rock, paper and scissors. The better quality data and the more data I gave the AI to learn from, the better it performed regardless of the complexity of the images.

In this example we used images as our domain. But deep learning can extend to many other domains, and the magic is that the underlying algorithm doesn’t change! The same algorithm can be used in speech recognition, food ingredients recognition and even taught to drive cars. Large companies like Google own vast quantities of data to train these AIs and thus stand at a competitive edge with their technology. Their data is ever expanding, as every time we email, text or use voice commands we provide richer data for them to improve their AI and serve their customers better.


A closer look into different types of machine learning

There is a distinct separation in machine learning, with supervised and unsupervised learning. Supervised learning is training our deep learning models on labelled data. That means for each image in my dataset, I have a label being cat, dog, fish etc. I am supervising the AI to learn a specific outcome. My dissertation project described above is an example of supervised learning, and many other popular applications of machine learning tend to fall in the same category.


Unsupervised/semi-supervised learning is where we remove the human supervision from the learning process entirely, or partly. It is harder to picture how an AI could just ‘learn’ from images without any guidance or with minimal guidance, but I will list a few examples. If we give an AI a car and a virtual environment where it can drive around, eventually with time it would learn to drive the car without crashing (it might take some time, but it will get there!). It does this by receiving a reward for its actions. May it be a positive reward for not crashing or a negative reward for doing so. Another example is learning patterns within our data that we as humans cannot see. Given many novels from different authors, AI could analyse and ‘cluster’ similar information together finding patterns in our data that only a machine could find. There are many other examples, so here are a few more.


How AND Digital could benefit from Machine Learning

Since joining AND Digital I have noticed a few domains where machine learning could be used to keep us on the cutting edge of technology. Our internal learning course booking system provides an interesting domain for deep learning to automate a pipeline, based on incoming requests and finding trends within our own company to help identify courses for both ANDis and clients. This could help us tend to preferences with less supervision and more expertise from AIs trained on previous knowledge, thus saving time and improving the growth of knowledge in our people.


Another example of a rich domain where deep learning or machine learning could be used is within Talk Talk. Talk Talk holds information of around four million customers across the UK. With such a rich data source, there are many problems and improvements that could use the hand of machine learning. Some examples are to spot relations between customers and customer packages that humans or simple analytics can not easily find.  These AIs are ever evolving and ever learning based on the accumulation of data, responding to changes in the market as well as user preferences.



Machine learning offers a large reward for our clients and internal projects, however there can be a large initial cost for implementing and designing the architectures to support this. In the long term this helps us stay on top of cutting edge technology and be of even larger interest to companies wanting to utilise the potential of deep learning. I feel it is crucial to start evaluating the potential areas to use these AIs to benefit our customers, as well as new improved solutions for our clients’ products.


Read More From This Author

Share this blog post

Related Articles


We’re looking for bright, dynamic people to join our team!

Discover More Roles