Differences

This shows you the differences between two versions of the page.

Link to this comparison view

iterative_teaching [2017/05/31 19:56] (current)
Line 1: Line 1:
 +https://​arxiv.org/​pdf/​1705.10470v1.pdf Iterative Machine Teaching
  
 +
 +In this paper, we consider the problem of machine
 +teaching, the inverse problem of machine
 +learning. Different from traditional machine
 +teaching which views the learners as batch algorithms,
 +we study a new paradigm where the
 +learner uses an iterative algorithm and a teacher
 +can feed examples sequentially and intelligently
 +based on the current performance of the learner.
 +We show that the teaching complexity in the iterative
 +case is very different from that in the batch
 +case. Instead of constructing a minimal training
 +set for learners, our iterative machine teaching
 +focuses on achieving fast convergence in the
 +learner model. Depending on the level of information
 +the teacher has from the learner model,
 +we design teaching algorithms which can provably
 +reduce the number of teaching examples and
 +achieve faster convergence than learning without
 +teachers. We also validate our theoretical findings
 +with extensive experiments on different data
 +distribution and real image datasets.