Interpreting the Residual Stream of ResNet18
This work provides incremental insights into the mechanistic understanding of DNN architectures, specifically for researchers in computer vision and neural network interpretability.
The study investigated the residual stream in ResNet18 to understand its role in visual object recognition, revealing that it acts as a flexible feature manager and facilitates scale invariant representations through mixtures of input and block features.
A mechanistic understanding of the computations learned by deep neural networks (DNNs) is far from complete. In the domain of visual object recognition, prior research has illuminated inner workings of InceptionV1, but DNNs with different architectures have remained largely unexplored. This work investigates ResNet18 with a particular focus on its residual stream, an architectural mechanism which InceptionV1 lacks. We observe that for a given block, channel features of the stream are updated along a spectrum: either the input feature skips to the output, the block feature overwrites the output, or the output is some mixture between the input and block features. Furthermore, we show that many residual stream channels compute scale invariant representations through a mixture of the input's smaller-scale feature with the block's larger-scale feature. This not only mounts evidence for the universality of scale equivariance, but also presents how the residual stream further implements scale invariance. Collectively, our results begin an interpretation of the residual stream in visual object recognition, finding it to be a flexible feature manager and a medium to build scale invariant representations.