Visualizing Deep Neural Networks for Speech Recognition with Learned Topographic Filter Maps
This addresses the challenge of interpreting internal structures in deep learning models for researchers and practitioners, though it is incremental as it adapts neuroscience techniques to a specific domain.
The paper tackled the problem of unintuitive visualization of deep neural network activations by training a convolutional speech recognition model with filters arranged in a 2D grid to create topographic maps, showing that this approach visualizes neuron activations more intuitively and groups phoneme-responsive neurons in specific regions.
The uninformative ordering of artificial neurons in Deep Neural Networks complicates visualizing activations in deeper layers. This is one reason why the internal structure of such models is very unintuitive. In neuroscience, activity of real brains can be visualized by highlighting active regions. Inspired by those techniques, we train a convolutional speech recognition model, where filters are arranged in a 2D grid and neighboring filters are similar to each other. We show, how those topographic filter maps visualize artificial neuron activations more intuitively. Moreover, we investigate, whether this causes phoneme-responsive neurons to be grouped in certain regions of the topographic map.