How to initialize your network? Robust initialization for WeightNorm & ResNets
Tipus de documentComunicació de congrés
Data publicació2019
Condicions d'accésAccés obert
Tots els drets reservats. Aquesta obra està protegida pels drets de propietat intel·lectual i
industrial corresponents. Sense perjudici de les exempcions legals existents, queda prohibida la seva
reproducció, distribució, comunicació pública o transformació sense l'autorització del titular dels drets
Abstract
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.
CitacióArpit, D.; Campos, V.; Bengio, Y. How to initialize your network? Robust initialization for WeightNorm & ResNets. A: Annual Conference on Neural Information Processing Systems. "Advances in Neural Information Processing Systems 32 (NIPS 2019)". 2019, p. 1-10.
Versió de l'editorhttp://papers.nips.cc/paper/9272-how-to-initialize-your-network-robust-initialization-for-weightnorm-resnets
Altres identificadorshttps://arxiv.org/abs/1906.02341
Fitxers | Descripció | Mida | Format | Visualitza |
---|---|---|---|---|
9272-how-to-ini ... for-weightnorm-resnets.pdf | 448,6Kb | Visualitza/Obre |