Tuesday, 17 April 2018

Deep learning creating a lot of hype and for good reason

Deep learning is a type of machine learning, based on a set of algorithms that model high-level abstractions in data, by using multiple processing layers with complex structures. Instead of organising data to run through predefined equations, deep learning sets up basic parameters about the data and trains the computer to learn on its own by recognising patterns using many layers of processing. This means it can train computers to perform human-like tasks, such as recognising speech, identifying images or making predictions.

Deep learning is already being used to make significant inroads into areas such as image recognition, fraud detection and the highly regulated credit risk modelling. In fact, SAS is currently working with credit bureau, Equifax, using deep learning techniques in credit risk modelling. The results are promising as the accuracy of the models has improved traditional techniques.

Organisations like Data Science Nigeria are working to speed up the application of machine learning to solve complex societal and business problems. Their approach is to get more people interested and skilled in data science so that Nigeria can become an outsourcing hub for international data science projects. They, too, have turned their focus to financial inclusion.

Bots that understand emotion 

Another exciting space in AI is bot technology. Chatbots are programmes that use natural language processing and AI to create conversations between machines and humans.

Instead of having a human respond to complaints or queries, this can now be done by a chatbot to save time and money on mundane and repetitive tasks. For example, responses to queries on bank accounts. Some banks are using bots to advise customers on financial advice and investments.

AI is teaching itself to think 

Until now, AI has generally been designed to do specific things like fraud detection. The human ability to perform tasks has always been greater than machines as we can generalise and perform a much wider set of functions.

But incredibly we're starting to see AI train itself to learn. 

In 2016, Google created a programme called AlphaGo. It was capable of beating even the most skilled human players at the ancient Chinese strategy game, Go - considered to be one of the most complicated games on earth.

But this was taken a step further through the creation of AlphaGo Zero, a programme provided with a very limited amount of training data. The idea was that it would learn by playing against itself. Over a period of time, AlphaGo Zero beat AlphaGo.

Essentially, it had taught itself to think. 

On the threshold of a future in which machines can think and learn - as we step into 2018 one could literally say nothing is impossible.

No comments:

Post a Comment