The Brain Behind the Machine
How machine learning is helping big data chart a picturesque new frontier
Without analytics, having access to big data is like possessing large volumes of books in a library – the information is dormant and insights are limited to what can be read on each page. Being oblivious of the context behind data, however, is now a thing of the past. Instead of stagnant libraries, developments in hardware and processing technology have allowed us to mine the petabytes of data collected everyday and process it almost like a thinking human brain.
This is where machine learning comes into the picture.
What is machine learning, and why is it useful?
By using algorithms to automate how analytic models are built – simply put, giving machines the ability to make sense of information in the context of the dataset – machine learning technology can be taught to look for unique insights and hidden relationships between data sources without being explicitly programmed on where to find them.
The algorithms behind this form of artificial intelligence have existed for decades. But the advent of new computing technologies has meant that machine learning today is not like the machine learning of years gone by. It is only recently that we have been able to develop hardware powerful enough to enable software to automatically apply complex mathematical calculations to big data over and over, and faster and faster each time.
Machine learning today finds applications in bioinformatics, medical diagnosis, natural language processing, robotics, sentiment analysis, speech recognition and stock market analysis. But can machine learning power all data analytics? After all, the majority of today’s data and analytics is text-based. Yet, pictures and videos take up most of the traffic on social media, most surveillance setups are video based and e-commerce relies heavily on how consumers respond to images.
Image and video data – a new language for machine learning?
To truly tap into the wealth of information contained in digital media worldwide, machine learning must break into the next frontier – image and video datasets. At Graymatics, we’ve already begun to discover what the machine’s “brain” is capable of when it starts to think and understand digital media data.
Our proprietary, industry-leading G3C (Graymatics Context Connect Cloud) platform relies upon machine vision, an image-focused subset of machine learning, to identify specific visual details and metadata through a smart tagging effort. We then use this metadata to process and understand how images and videos within a dataset relate to one another, packaging these insights into solutions utilised by leaders in the telecom, e-commerce and surveillance sectors.
The fuel for our G3C engine is an image dataset of more than a million images, cultivated across several various different categories. At Graymatics, we utilise multiple techniques – including Bag-of-Words representation, spatial pyramid matching, topic models and sparse coding – to train our algorithm to draw associations between multiple images, as well as to identify the unique visual characteristics of each image. The trained algorithm is then applied to every new image transferred on our cloud infrastructure. These techniques help give us a minimal error rate against the most state-of-the-art standard available.
Our algorithm can also begin to map out relationships between these images and videos to identify unique insights. For instance, instead of only using “dumb” search methods involving searching via specific filters, Graymatics offers users the option of dynamic smart searching. For instance, a user may want to search for media relating to “fruit” through our platform. Instead of only showing images which have been manually tagged “fruit”, our machine learning technology can generate results from visual content (images or video scenes depicting fruits), or even speech content (the moment the word “fruit” is mentioned).
Powered by machine learning, we can finally unlock the treasure trove of insights image and video data has to offer. In surveillance, machine learning technologies can employ facial recognition, vehicle classification and object tracking to rapidly scan CCTV videos to detect or prevent crime and terrorism. In e-commerce, retailers are using this technology to broaden search capabilities to coax buyers towards the products they really want. In consumer-facing companies, marketing teams can tap into social media posts to build up their knowledge of customers, consumers and trends in real time.
Machine learning has countless useful applications across a range of sectors, and its proliferation marks the start of a new digital revolution. As image recognition technology gets smarter and more sophisticated, this opens up a new, automated data ecosystem able to intuit and analyse patterns that might otherwise have been missed. The challenge now is for businesses to leverage the visual insights at their fingertips to help big data chart a picturesque new frontier.