Learning By Asking, CVPR. 2018

Related Works

Exploratory Learning

  • An agent explores the environment
  • Usage: computer games, navigation, multi-user games, inverse kinematics, motion planning for humanoids
  • It operates by reinforcement learning in which a delayed reward which is used to learn a policy that maximizes the expected rewards
  • LBA does not have a "sparse delayed reward" system
  • LBA closely resembles Contextual multi-armed bandits

Learning By Asking

Variables

  • I: images
  • Q: a set of all possible questions
  • A: a set of all possible answers
  • N: number of images
  • Dtrain: {I1, ...., IN}

Training Tyime

  1. I: images
  2. Q: a set of all possible questions
  3. A: a set of all possible answers
  4. N: number of images
  5. Dtrain: {I1, ...., IN}