概念
rl training
创建2026-04-15
更新2026-04-15
阅读量级1 分钟
概念导读

Interactive Imitation Learning extends standard imitation-learning by allowing the learner to query the expert interactively during training, rather than learning only from a fixed dataset of demonstrations. This addresses the exposure-bias problem by ensuring the learner receives guidance on states it actually encounters during rollout, similar to how on-po...

  1. Related
  2. References

Interactive Imitation Learning

Interactive Imitation Learning extends standard imitation-learning by allowing the learner to query the expert interactively during training, rather than learning only from a fixed dataset of demonstrations. This addresses the exposure-bias problem by ensuring the learner receives guidance on states it actually encounters during rollout, similar to how on-policy methods in distillation address distribution mismatch.

  • exposure-bias — Interactive IL directly addresses exposure bias by querying experts on learner-visited states
  • imitation-learning — Broader imitation learning paradigm
  • on-policy-distillation — OPD similarly addresses distribution mismatch through on-policy rollouts

References