deletions | additions       diff --git a/textbf_Considerations_The_goal_of__.tex b/textbf_Considerations_The_goal_of__.tex  index 295c03e..fc565ab 100644  --- a/textbf_Considerations_The_goal_of__.tex  +++ b/textbf_Considerations_The_goal_of__.tex 
...
The goal of the above HOG scheme is the efficient extraction and detection of features. There is an obvious difference with the convolutional layers of a CNN, which instead learn the features from the data.  One of the goals of quickly learnable part detectors was to allow on-the-fly creation for temporary use.  A problem with neural nets, similar to gmms is the permutability of the solutions due to the sums.