3/6 Method
The pipeline runs in three steps.
• Score each sim demo against target rollouts. A base policy is rolled out in the target domain, each rollout carrying its own return. For each demo, an influence function asks: if this demo had more weight, which rollouts would https://t.co/dk7efn2noz
The pipeline runs in three steps.
• Score each sim demo against target rollouts. A base policy is rolled out in the target domain, each rollout carrying its own return. For each demo, an influence function asks: if this demo had more weight, which rollouts would https://t.co/dk7efn2noz