We are interested in achieving the best possible recognition rates for
hand-written mathematics.   This problem has challenges that go beyond
the usual natural language handwriting recognition:  multiple alphabets
are used, the number of single-stroke and few-stroke symbols numbers
in the hundreds, writers form symbols from some of the alphabets in
idiosyncratic non-standard manners, layout is two dimensional, and
there is no fixed dictionary of words to aid in disambiguation.

While writer-independent handwriting recognition systems are now achieving
good recognition rates, writer-dependent systems will always do better.
We expect this difference in performance to be even larger for handwriting
math systems.  In the past, it would not be too inconvenient for
a writer to train a system on his or her computer.  Today, however,
each user will typically have multiple devices used in different settings,
or even simultaneously.

We present a method to share training data among devices and,
as a side benefit, to collect user corrections over time to
improve personal writing recognition.  This is done with the aid
of a handwriting profile server to which various handwriting applications
connect, reference and update.   The user's handwriting profile consists of
a cloud of sample points, each representing one character in a functional basis.
This provides compact storage on the server, rapid recognition on the client,
and support for handwriting neatening.