Deep Learning Applications
Deep learning applications are present in a wide area of business as well as scientific domains and take advantage of big data today. Examples are from the domains of computer vision, automatic speech recognition, natural language processing, remote sensing, face recognition, bioinformatics, neuroscience, genomics, ore more generally image registration and microscopy. This article lists some of them with further reading material below. Another article provides more general understanding about deep learning that we recommend to read before.
Interesting is that deep learning models learn multiple levels of representation and abstraction that automatically help to make sense of various datasets such as images, sound, or text. Therefore one application is even before modeling data with other models like Support Vector Machines (SVM). This means to create automatically better data representations and create deep learning models to learn these data representations from large-scale unlabeled data before using traditional machine learning models.
Remote sensing is an application domain with hyper-spectral or multi-spectral datasets where labeled data in form of ‘groundtruth’ is rather rare. In the field of remote sensing one can use a pre-trained convolutional neural network (CNN) that was trained on the ImageNet challenge data in order to improve the classification of remote sensing data. This method is also referred to as transfer learning. In other words the CNN is first trained on a completely different classification problem in order to extract an initial set of representations from the rather simple images. In the next step the derived representations are transferred into a supervised CNN classification problem including class labels of remote sensing data. It is a two-stage approach that successfully deals with limited data within remote sensing and achieves a high classification accuracy. More details are available here.
Another interesting application field is scientific image registration and microscopy in the context of the imaging flow cytometry that combines the fluorescence sensitivity and high-throughput capabilities of flow cytometry with single-cell imaging.. The results are big datasets ideal to be used with deep learning algorithms. The approach one can take is to combine convolutional neural network (CNN) networks with non-linear dimension reduction. The goal is that learned features of the network are used to visualize, organize, and biologically interpret single-cell data. More details are available here.
Details on Deep Learning Applications
We recommend to watch the following video about this subject:
