diff --git a/fastai_dl_terms.md b/fastai_dl_terms.md
index 0b82612..47a31bc 100644
--- a/fastai_dl_terms.md
+++ b/fastai_dl_terms.md
@@ -5,11 +5,12 @@
 `arch` architecture (such as resnet18, resnet34, resnext50)  
 `aug_tfms = transforms_side_on` data augmentation using flipping image on its side (horizontal, flip right to left)  
 `aug_tfms = transforms_top_down` data augmentation (flipping image top down)  
-`learn.bn_freeze`  Batch Norm Unfreeze
+`do_scale=True`  scaling - neural nets really like input data to be ~N(0, 1)  
 `bptt` back propagation through time   
 `bs` = batch size  
 `.cuda()` we tell it manually to use the (default number of) GPUs    
-`.cuda(2)` specify number of GPUs to use is 2  
+`.cuda(2)` specify number of GPUs to use is 2   
+`learn.bn_freeze`  Batch Norm Unfreeze
 `lo` = layer optimizer  
 `lr` learning rate  
 `lr_find()` learning rate finder