Permalink
Find file
Fetching contributors…
Cannot retrieve contributors at this time
18 lines (12 sloc) 490 Bytes
layout permalink
page
/convnet-tips/

Addressing Overfitting

Data Augmentation

  • Flip the training images over x-axis
  • Sample random crops / scales in the original image
  • Jitter the colors

Dropout

  • Dropout is just as effective for Conv layers. Usually people apply less dropout right before early conv layers since there are not that many parameters there compared to later stages of the network (e.g. the fully connected layers).