Brandon Paulson
Homework 2 – “Specifying Gestures by Example”

Discussion:

	This paper was very easy to follow and made a lot of sense to me.  
The one aspect of Rubine’s method that I found the most interesting was 
creating a feature vector and using weights on that vector as a way of 
training.  I think this leaves a lot of flexibility because features can 
be added or removed from the vector to achieve different results.  A lot 
of work could be done to discover other features that can be used to 
improve training and recognition.  Another thing I found interesting was 
how gestures could be recognized in real-time and before the user actually 
finishes the gesture.  I think this is a really nice feature as long as it 
doesn’t misclassify the gesture.  It would be very annoying to be attempting 
to draw one thing but the recognizer thinks you are drawing something 
different and it enters it’s “manipulation” mode.  The main downfall of the 
method (in my opinion) is that it is only a single-stroke recognizer.  Most 
users probably wouldn’t write letters and symbols using just single-stokes.  
I noticed a few other minor details in the paper.  For example, GDP has a 
“rotate-scale” gesture.  It would seem more logical to make these two 
separate gestures (although I guess it’s not a very big deal since GDP was
only meant to show the capability of GRANDMA).  Another thing I thought was 
odd was that the author removed input points that are within 3 pixels of the 
previous input point.  It would seem more logical to me to make this a 
percentage of pixels (based on resolution) instead of a fixed number of pixels.