Unverified50% confidenceFactExact time
The paper categorizes existing OPSD approaches into two types: one where the teacher is fixed throughout training and has access to the full reference; the other where the teacher is unchanged within each response but adaptively varies across responses in the degree of exposure to partial reference.
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe paper "Every Token Needs Its Own Teacher: Multi-teacher Adaptive On-policy Self-distillation" points out that existing OPSD methods assume the teacher is fixed, and this assumption has fundamental limitations.74% similarUnverifiedMTA-OPSD transforms the teacher selection problem in self-distillation into a token-level dynamic routing problem, where the teacher is no longer a static hyperparameter but a variable that changes with training state and token context.72% similarUnverifiedSteered-Teacher OPSD (ST-OPSD) simultaneously mitigates both failure modes of reference hallucination and reconsideration suppression through activation steering techniques.71% similarUnverifiedPIVOT decomposes teacher guidance into two complementary mechanisms: trajectory-level intervention (inserting teacher corrections into student trajectories) and token-level supervision (providing dense target signals for the correction segments).68% similarUnverifiedOPRD's teacher and student share the same two-stage initialization: first Coding SFT, then Instruction-Following SFT.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1115269API
curl https://kongchang.com/api/v1/knowledge/claims/1115269MCP
get_claim(id=1115269)