Weighted Residuals for Very Deep Networks
This addresses a specific problem in computer vision for researchers and practitioners by enabling more efficient training of very deep networks, though it is incremental over existing residual methods.
The paper tackles the convergence and accuracy issues of very deep residual networks by introducing weighted residuals, which improve compatibility with ReLU and initialization, resulting in faster convergence and 95.3% accuracy on CIFAR-10 with a 1192-layer model.
Deep residual networks have recently shown appealing performance on many challenging computer vision tasks. However, the original residual structure still has some defects making it difficult to converge on very deep networks. In this paper, we introduce a weighted residual network to address the incompatibility between \texttt{ReLU} and element-wise addition and the deep network initialization problem. The weighted residual network is able to learn to combine residuals from different layers effectively and efficiently. The proposed models enjoy a consistent improvement over accuracy and convergence with increasing depths from 100+ layers to 1000+ layers. Besides, the weighted residual networks have little more computation and GPU memory burden than the original residual networks. The networks are optimized by projected stochastic gradient descent. Experiments on CIFAR-10 have shown that our algorithm has a \emph{faster convergence speed} than the original residual networks and reaches a \emph{high accuracy} at 95.3\% with a 1192-layer model.