Our world is full of devices and applications that rapidly generate, store, and transmit huge amounts of data. Due to these developments, enormous quantities of data have become available for all kinds of analyses. For example, social media data, self-driving cars filled with sensors, intelligent home and office equipment, internet and browsing behaviour, digital camera images, smartphone apps, and other wearables. Increasingly, these so-called raw data cannot completely be classified or labelled and are usually not easy to analyze with established statistical analysis techniques.
These fast growing volumes and varieties of available data, cheaper and more powerful computational processing, and affordable data storage has recently created huge interest in machine and deep learning. The foundation of these techniques is the use of pattern recognition on complex and multispectral data. By making software autonomous or using iterative feedback to discover associations in data, patterns are found and anomalies detected. This leads to a taxonomy for finding clusters and classifications.
Deep learning algorithms can learn discriminative features from large sets of data which is both systematic and automatic. Thus, within deep learning a transition from manual feature engineering to unsupervised feature learning is made. Deep learning is playing an instrumental role in analysis of huge volumes of multidimensional and complex data, like data collected from sensors. We have experience with applying deep neural networks on terabytes of triaxial accelerometer data for human activity recognition at both algorithmic and infrastructure levels.