Beyond Signal Propagation: Is Feature Diversity Necessary in Deep Neural Network Initialization?
This challenges a common belief in deep learning initialization, potentially simplifying training procedures, but it is incremental as it builds on existing signal propagation theories.
The authors tackled the problem of whether feature diversity is necessary in neural network initialization by constructing a deep convolutional network with identical features, initialized with near-zero weights, that still achieved high accuracy on standard benchmarks, indicating random diverse initializations are not required.
Deep neural networks are typically initialized with random weights, with variances chosen to facilitate signal propagation and stable gradients. It is also believed that diversity of features is an important property of these initializations. We construct a deep convolutional network with identical features by initializing almost all the weights to $0$. The architecture also enables perfect signal propagation and stable gradients, and achieves high accuracy on standard benchmarks. This indicates that random, diverse initializations are \textit{not} necessary for training neural networks. An essential element in training this network is a mechanism of symmetry breaking; we study this phenomenon and find that standard GPU operations, which are non-deterministic, can serve as a sufficient source of symmetry breaking to enable training.