In high-dimensional linear and one-step feature-learning models, a regularized student can outperform its teacher by fixing under-regularization, using better regularization structure, or retaining pretrained hard features.
Similarly, for the remaining terms, we have d−1 X Tr h ˆR ˜R i →P mt,1ms,1 d−1 X Tr h ˆR ˜R2 i →P mt,1ms,2, and d−1 X Tr h ˆR2 ˜R i →P mt,2ms,1
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
On the Mechanisms of Weak-to-Strong Generalization: A Theoretical Perspective
In high-dimensional linear and one-step feature-learning models, a regularized student can outperform its teacher by fixing under-regularization, using better regularization structure, or retaining pretrained hard features.