learning in abstact domain