How to Initialize your Network? Robust Initialization for WeightNorm & ResNets
This work addresses a critical bottleneck in deep learning by improving initialization for weight-normalized and residual networks, which is incremental but important for enhancing training stability and performance in these widely used architectures.
The authors tackled the problem of parameter initialization for weight-normalized and residual networks, proposing a novel strategy that prevents information explosion or vanishing across layers. Their method, validated through over 2,500 experiments on image datasets, outperformed existing initialization techniques in generalization, robustness, and variance, especially for deeper networks where prior methods failed to train.
Residual networks (ResNet) and weight normalization play an important role in various deep learning applications. However, parameter initialization strategies have not been studied previously for weight normalized networks and, in practice, initialization methods designed for un-normalized networks are used as a proxy. Similarly, initialization for ResNets have also been studied for un-normalized networks and often under simplified settings ignoring the shortcut connection. To address these issues, we propose a novel parameter initialization strategy that avoids explosion/vanishment of information across layers for weight normalized networks with and without residual connections. The proposed strategy is based on a theoretical analysis using mean field approximation. We run over 2,500 experiments and evaluate our proposal on image datasets showing that the proposed initialization outperforms existing initialization methods in terms of generalization performance, robustness to hyper-parameter values and variance between seeds, especially when networks get deeper in which case existing methods fail to even start training. Finally, we show that using our initialization in conjunction with learning rate warmup is able to reduce the gap between the performance of weight normalized and batch normalized networks.