PCACE: A Statistical Approach to Ranking Neurons for CNN Interpretability

In this paper we introduce a new problem within the growing literature of interpretability for convolution neural networks (CNNs). While previous work has focused on the question of how to visually interpret CNNs, we ask what it is that we care to interpret, that is, which layers and neurons are worth our attention? Due to the vast size of modern deep learning network architectures, automated, quantitative methods are needed to rank the relative importance of neurons so as to provide an answer to this question. We present a new statistical method for ranking the hidden neurons in any convolutional layer of a network. We define importance as the maximal correlation between the activation maps and the class score. We provide different ways in which this method can be used for visualization purposes with MNIST and ImageNet, and show a real-world application of our method to air pollution prediction with street-level images.

Evaluating Word Embeddings with Categorical Modularity

We introduce categorical modularity, a novel low-resource intrinsic metric to evaluate word embedding quality. Categorical modularity is a graph modularity metric based on the $k$-nearest neighbor graph constructed with embedding vectors of words from a fixed set of semantic categories, in which the goal is to measure the proportion of words that have nearest neighbors within the same categories. We use a core set of 500 words belonging to 59 neurobiologically motivated semantic categories in 29 languages and analyze three word embedding models per language (FastText, MUSE, and subs2vec). We find moderate to strong positive correlations between categorical modularity and performance on the monolingual tasks of sentiment analysis and word similarity calculation and on the cross-lingual task of bilingual lexicon induction both to and from English. Overall, we suggest that categorical modularity provides non-trivial predictive information about downstream task performance, with breakdowns of correlations by model suggesting some meta-predictive properties about semantic information loss as well.