深度卷积神经网络的增量式训练

Mar, 2018

Incremental Training of Deep Convolutional Neural Networks

Roxana Istrate, Adelmo Cristiano Innocenza Malossi, Costas Bekas, Dimitrios Nikolopoulos

TL;DR提出了一种增量训练方法，将原始网络分成子网络，并在训练过程中逐步将其合并到运行的网络中。此方法通过引入前瞻初始化，使网络动态生长更平滑，并可用于仅使用全局参数分数的情况下识别原始最先进网络的较小分区，以实现更快的训练。在CIFAR-10上报告了ResNet和VGGNet的训练结果。

Abstract

We propose an incremental training method that partitions the original network into sub-networks, which are then gradually incorporated in the running network during the training process. To allow for a smooth