Conceptdivergence tokens
Disagreement Token
A token whose frequency or distribution differs markedly between harm-accepting and harm-avoiding teacher responses. Such tokens are regarded as key carriers of behavioral transfer in knowledge distillation.
Timeline (last 90 days)
Oct 14
Excluding a sparse set of "divergence tokens" can significantly reduce the imprinting effect, suggesting that certain specific words or expression patterns in the teacher's responses are the key carriers of behavioral transfer.
Unverified50%