Object recognition in images, automatic classification, learning a language, detection of signatures… little by little we are beginning to solve problems that until now were conceptually too complicated for a computer and that only a person was able to reason out. Artificial intelligence has been able to go one step further to emulate the learning of our brain through convolutional neural networks, revolutionizing the world of automatic learning. But what exactly are they? And most importantly, what possibilities do they offer us to improve conventional documentary systems?

As their name implies, artificial neural networks mimic the functioning of neural networks in living organisms, so that a set of elements (neurons) connected to each other work together, without there being a specific task for each one. Gradually, through experience, neurons are creating and reinforcing certain connections to “learn” a series of more significant elements about whatever they are shown.

But how does a neural network “learn” the logic in unstructured information? Learning with convolutional networks is based on searching for common characteristics in small groups of inputs, e.g. in the case of images of scanned documents, of pixels. These small groups may be more or less homogeneous edges or colours. In addition, we always seek to detect the same characteristic in all groups of images, so we can train the network to gradually improve its results with experience, just as a human would.

Big technology companies like Google are already using this type of neural network to improve their applications. For example, in Street View a convolutional neural network achieved 96% accuracy when recognizing street numbers in the images taken by their cars. Convolutional networks have also been used to improve Android voice recognition or to save electricity in their data centres.

The advantages of applying these advanced systems of convolutional networks to traditional document management systems are obvious: what could only be done by people before, can now be done by machines, from automated classification, data capture and signature detection in contracts. With an automated document processing system such as TAAD, your company will be able to improve productivity exponentially in manual processes, possibly extending to document problems that were previously impossible to manage due to the large volume. A clear example has been the Barclays project for Caixabank, where the use of an automated classification system has allowed us to bring to life a project that would have been impossible to carry out in terms of time and cost through traditional document processing.