Paper summary
Paper summary: DAPD: Dual-Anchored Policy Distillation
On-policy self-distillation with a privileged teacher can cause "privilege illusion": the student acts as if it still had information it can't see at inference. DAPD anchors training along matched-information paths to prevent this.
- If the teacher sees privileged information, the student may learn behaviour that only makes sense with that information, which it won't have later.
- The fix aligns teacher and student along paths where both have the same information.