Multi-group learners in the transductive setting incur error penalties that can grow linearly with the number of groups (up to ~sqrt(n)), in contrast to at most logarithmic penalties independent of group count in the group-realizable statistical setting.
Tibshirani
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
fields
cs.LG 2years
2026 2verdicts
UNVERDICTED 2representative citing papers
Multicalibration has minimax sample complexity Θ̃(ε^{-3}) when the number of groups is at most ε^{-κ} for fixed κ>0, versus Θ̃(ε^{-2}) for marginal calibration, with a sharp threshold at κ=0.
citing papers explorer
-
The price of multi-group transductive learning
Multi-group learners in the transductive setting incur error penalties that can grow linearly with the number of groups (up to ~sqrt(n)), in contrast to at most logarithmic penalties independent of group count in the group-realizable statistical setting.
-
The Sample Complexity of Multicalibration
Multicalibration has minimax sample complexity Θ̃(ε^{-3}) when the number of groups is at most ε^{-κ} for fixed κ>0, versus Θ̃(ε^{-2}) for marginal calibration, with a sharp threshold at κ=0.