EasyDeepLearn

Curriculum learning in RL — how to design one?

medium

Answer

  • Order training from easy to hard tasks: (1) start with dense reward, gradually sparsify.
  • (2) Start with easy env, add complexity (obstacles, distractors).
  • (3) Adaptive: automatic difficulty tuning to keep learning at frontier of ability (POET, ADR).
  • (4) Self-play induces natural curriculum via opponent's growing skill.
  • (5) Sub-goal decomposition: reach intermediate states before final.
  • Modern: adversarial curriculum where env teacher proposes tasks the student can barely solve.
Check yourself — multiple choice
  • Random
  • Order easy→hard: dense→sparse reward, easy→complex env, adaptive difficulty (POET/ADR), self-play natural curriculum, sub-goal decomposition, adversarial teacher-student — many designs
  • Just random
  • Not real

Curriculum: dense→sparse / easy→complex / adaptive (POET/ADR) / adversarial teacher.

#policy-methods#engineering

Practise Reinforcement Learning

214 interview questions in this topic.

Related questions