MixSD mixes tokens from the base model's expert and naive conditionals to create distribution-aligned supervision for knowledge injection, yielding better memorization-retention trade-offs than SFT across scales and benchmarks.
Be Your Own Teacher: Improve the Performance of Convolutional Neural Networks via Self Distillation , year=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
MixSD: Mixed Contextual Self-Distillation for Knowledge Injection
MixSD mixes tokens from the base model's expert and naive conditionals to create distribution-aligned supervision for knowledge injection, yielding better memorization-retention trade-offs than SFT across scales and benchmarks.