Channel redundancy and overlap in convolutional neural networks with channel-wise NNK graphs
This provides insights into CNN representations for researchers, but it is incremental as it builds on existing graph-based analysis methods.
The paper tackled the problem of interpreting high-dimensional feature spaces in CNNs by analyzing channel-wise NNK graphs to quantify channel overlap and redundancy, finding that redundancy varies with layer depth and regularization and correlates with generalization performance.
Feature spaces in the deep layers of convolutional neural networks (CNNs) are often very high-dimensional and difficult to interpret. However, convolutional layers consist of multiple channels that are activated by different types of inputs, which suggests that more insights may be gained by studying the channels and how they relate to each other. In this paper, we first analyze theoretically channel-wise non-negative kernel (CW-NNK) regression graphs, which allow us to quantify the overlap between channels and, indirectly, the intrinsic dimension of the data representation manifold. We find that redundancy between channels is significant and varies with the layer depth and the level of regularization during training. Additionally, we observe that there is a correlation between channel overlap in the last convolutional layer and generalization performance. Our experimental results demonstrate that these techniques can lead to a better understanding of deep representations.