Computer Science

Imagination Sampling

  • Eric Holloway


Machine Learning, despite its name, can incorporate an oracle.  One common form of oracle interaction is known as active learning.
Active learning samples ({x,y}) from an oracle for f (the function to be learned).  Imagination sampling is the converse of active learning.  Imagination sampling asks an oracle for hypotheses h from H (hypothesis space).  In this paper imagination sampling is compared with a purely algorithmic approach to determine if oracle interaction outperforms a purely algorithmic approach.  The theoretical basis for imagination sampling is developed and illustrated by simulating an oracle.