Deep Neural Networks (DNNs) have demonstrated impressive performance in complex machine learning tasks such as image classification, human action recognition or sentiment analysis. These methods are so successful because they are forming an efficient internal representation of the learning problem by applying a sequence of nonlinear transformations to the input. This provides a very efficient way to fight the curse of dimensionality and learn the optimal features for a given problem.
We develop novel deep architectures for a variety of complex real-world tasks and efficiently train the network using GPU’s.
The main focus of our work lies on
- Training and evaluating convolutional networks for image & video annotation tasks.
- Optimally combining different types of information (e.g., spatial & temporal).
- Understanding and interpreting the reasoning of the system.
- Mimicking and improving well-known algorithms by deep learning.
- S. Bach, A. Binder, G. Montavon, F. Klauschen, K.-R. Müller, and W. Samek, “On Pixel-wise Explanations for Non-Linear Classifier Decisions by Layer-wise Relevance Propagation”, PLOS ONE, vol. 10, no. 7, pp. e0130140, July 2015.