Tube Loss is a novel loss function enabling simultaneous prediction interval bound estimation with asymptotic coverage guarantees, tunable positioning for skewed distributions, and trade-offs between coverage and width via single optimization.
George Chryssolouris, Moshin Lee, and Alvin Ramsey
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Tube Loss: A Novel Approach for Prediction Interval Estimation
Tube Loss is a novel loss function enabling simultaneous prediction interval bound estimation with asymptotic coverage guarantees, tunable positioning for skewed distributions, and trade-offs between coverage and width via single optimization.