Deep Learning (Cognitive Computing)

Penetrating Data with Artificial Neural Networks

We are currently experiencing an artificial intelligence revolution thanks to Deep Learning. Deep Learning is a new information processing method based on artificial neural networks which has led to ground-breaking achievements in image recognition, speech processing and robotics.

Central to its success is the fact that Deep Learning is most effective when particularly large amounts of data – Big Data – are available to train the neural networks. Thanks to its many years’ experience working on neurocomputing and Big Data Analytics solutions the Fraunhofer IAIS in Germany is among the pioneering developers of Deep Learning methods for industry.

We have successfully implemented Deep Learning processes in projects as diverse as Intelligent Automotive, Multimedia or Customer Churn Prediction.

Applications

Multimedia – Automatic Speech Recognition

Despite major advances in automatic speech recognition some circumstances such as loud noises in cars, strong dialects and animated discussions are still too much of a test for speech recognition algorithms. Our scientists are currently investigating how the remaining challenges can be overcome with the help of deep neural networks. 

Pattern Recognition in Geological Data

Deep Learning techniques help aid the search for interesting seismic data structures: Our researchers train deep neural networks using real data from the oil and gas industries to be able to search for structures and configurations in seismic data clusters. In doing so they help geologists find invaluable raw material deposits more quickly and with greater certainty using hardware specifically designed for training and implementing deep neural networks. 

Automotive – Deep Learning for Object Recognition

Standard traffic sign recognition systems can only reliably recognize round traffic signs. Frequently, however, in areas such as near building sites other shapes of sign are used which could not until recently be recognized by automated systems or at least not recognized to a sufficient degree. Our researchers have dedicated their efforts to developing innovative solutions based on Deep Learning neural networks which can be integrated with existing automated recognition systems.