Allan,

For time-series, the path map gives the MAP estimate of a hidden
continuous-valued series variable X given some sequence of observations
Y.   X can evolve nonlinearly and Y can be a nonlinear function of X. 
It's fully learnable.

Examples in 

"Voice Puppetry"  Proc SIGGRAPH99  (X=3D facial motion; Y=voice)
"Shadow Puppetry" Proc ICCV99      (X=3D body motion; Y=shadows)

There's a web-available summary at
http://www.cis.ohio-state.edu/~szhu/workshop/Brand.html

cheer,
Matt Brand

Reply via email to