4

I found the phrase "high-capacity cnn" in these two papers:

1.Rich feature hierarchies for accurate object detection and semantic segmentation

2.Region-based Convolutional Networks for Accurate Object Detection and Segmentation

I've searched it up on google but I can't seem to find a good one.

Shieh Kai Yong
  • 43
  • 6

1 Answers1

3

This usually refers to the complexity of the underlying pattern that the neural network is able to learn. Usually going deep increases the capacity of the neural network by increasing the number of model parameters which means it can fit more complex functions. However, increasing capacity by going deep also increases your chance of over-fitting and issues such as regularisation becomes important. It usually also means you need to increase the number of training samples.

Pankaj Daga
  • 709
  • 5
  • 15
  • So, is this has the same meaning as "large cnn model"? – Shieh Kai Yong Oct 31 '16 at 13:40
  • 1
    This term is abused and used quite loosely but typically the capacity refers to the capacity to explore the function space. This is usually correlated with the depth of the NN. So usually `deep` neural networks are high capacity. If by `large`, you mean a large number of layers, than yes. – Pankaj Daga Oct 31 '16 at 14:04
  • 2
    I suppose that Residual Nets is the way how one should go deeper with increasing performance. – cagatayodabasi Oct 31 '16 at 17:56