Abstract: Deep Learning is an exploding area of machine learning based on data representations using multiple levels of abstraction. Deep neural network algorithms have recently obtained state of the art results for classification of large data sets due to advancement in computing power and the development of new techniques. Other strategies for data representation and feature extraction, such as topic modeling based strategies have also recently progressed. Topic models combine data modeling with optimization to learn hidden thematic structures in data. We propose a novel approach that combines the interpretability and predictability of topic modeling learned representations with the robust classification attributes of deep neural networks, introducing a deep nonnegative matrix factorization (deep NMF) framework capable of producing reliable, interpretable, and predictable hierarchical classification of text, audio, image and high dimensional data, far exceeding existing approaches.