Curriculum learning in RL — how to design one?
mediumAnswer
- 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.