The Tunnel Effect: Building Data Representations in Deep Neural Networks
This reveals a structural issue in deep networks affecting generalization, which is incremental but has implications for continual learning.
The paper demonstrates that deep neural networks for image classification split into two parts: initial layers create linearly-separable representations, while later layers (the tunnel) compress them with minimal performance impact, and it shows the tunnel degrades out-of-distribution generalization.
Deep neural networks are widely known for their remarkable effectiveness across various tasks, with the consensus that deeper networks implicitly learn more complex data representations. This paper shows that sufficiently deep networks trained for supervised image classification split into two distinct parts that contribute to the resulting data representations differently. The initial layers create linearly-separable representations, while the subsequent layers, which we refer to as \textit{the tunnel}, compress these representations and have a minimal impact on the overall performance. We explore the tunnel's behavior through comprehensive empirical studies, highlighting that it emerges early in the training process. Its depth depends on the relation between the network's capacity and task complexity. Furthermore, we show that the tunnel degrades out-of-distribution generalization and discuss its implications for continual learning.