A New Pruning Method to Train Deep Neural Networks
IEEE internet of things journal(2018)
摘要
Deep neural networks are very powerful models for machine learning tasks. However, suffering from overfitting and gradient vanishing problems, they are difficult to train. We proposed a method of gradually pruning the weakly connected weights to train deep neural networks and an effective strategy to identify the weak connections. Our method can improve the conventional stochastic gradient descent and can get even better performance than the widely used dropout method for deeper models.
更多查看译文
关键词
Deep neural networks,Overfitting,Stochastic gradient descent
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要