Multifaceted Feature Visualization: Uncovering the Different Types of Features Learned By Each Neuron in Deep Neural Networks

We can better understand deep neural networks by identifying which features of their neurons have learned to detect. To do so, researchers have Deep Visualization techniques including activation maximization, which generates inputs (e. g. images) that maximally activate each. A limitation of current techniques is that they assume each neuron only one type of feature, but we know that neurons can be multifaceted, that they fire in response to many different types of features: for example, grocery store class neuron must activate either for rows of produce or for a. Previous activation maximization techniques constructed images regard for the multiple different facets of a neuron, creating mixes of colors, parts of objects, scales, orientations, etc. , we introduce an algorithm that explicitly uncovers the multiple facets of neuron by producing a synthetic visualization of each of the types of that activate a neuron. We also introduce regularization methods that state-of-the-art results in terms of the interpretability of images by activation maximization. By separately synthesizing each type of a neuron fires in response to, the visualizations have more appropriate and coherent global structure. Multifaceted feature visualization thus a clearer and more comprehensive description of the role of each.