Getting started with deep learning

Categories: Machine Learning |

Getting started with deep learning

In Belatrix’s recent machine learning survey, we found that nearly 60% of organizations are investigating or have already started implementing machine learning and artificial intelligence. However, while there is a lot of interest, many are struggling with how to get started. There is a lot of media attention on the topic, but little in the way of concrete guidance for organizations who want to dip their feet into it and see what they can do.

As a result Belatrix has just published a new whitepaper, “Getting started with deep learning”. Deep learning, or neural networks are one of the key subtopics of machine learning. When we talk about deep learning, we are implicitly referring to neural networks which are a representation of our neurons and how they work. Neural networks have been used to solve different problems ranging from natural language processing, computer vision, sentiment analysis, voice recognition, and autonomous vehicles. Common consumer applications, such as Google Photos, rely on deep learning to help people find the photo they are searching for (Google even trained a deep learning machine to be able to work out the location of almost any photograph).

As a result, the whitepaper outlines the reasons why you should be interested in deep learning, as well as some of the challenges – such as how can you overcome the “black box” of AI. We explain how you can use platforms, such as Turi GraphLab, H2O, BisQue, or Google’s Cloud Machine Learning Platform to get started. Each of these platforms have their own specific benefits and drawbacks. There are also a variety of different neural nets that you can use – for example, for text processing or sentiment analysis you could consider using a recurrent neural network (RNN), while for image processing it’s better to consider a convolutional neural network (CNN or ConvNet).

Finally the report provides a series of use cases where deep learning can be applied. In FinTech, enterprises are already using neural networks and other machine learning techniques in areas such as fraud detection, customer service, and marketing. Fraud detection is particularly interesting example, because there are large amounts of unstructured data. PayPal for instance uses the H20 platform I mentioned earlier, to analyze fraud protection at the transaction, account, and network levels. With deep learning they are able to analyze and detect the increasingly complex patterns which characterize today’s online fraud.

Here at Belatrix meanwhile, we’ve been using neural networks to help determine the risk or likelihood of someone leaving the company. We’re proud that Belatrix has one of the lowest attrition rates in the technology services industry. By using neural networks we can better analyze the factors which may lead to someone wanting to leave, and proactively address them. We’re doing this based on data which we already have in our existing HR, project management and training systems. As a result, deep learning is making us a better organization. And this is really the key to getting started with machine learning and neural networks. When we talk to customers just starting their journey, we work with them to identify a particular niche which could be improved. For example, can we develop a particular function which will delight their customers? This provides significant return on limited investment.

As our survey revealed, machine learning will be a key priority for organizations in 2017, helped by the fact that the technology is maturing rapidly. 81% of respondents believed it will have some or significant impact on their organization within the next 5 years. We’re excited about seeing more and more use cases emerging, and working with our customers on their journeys.

 

Leave a comment