Bayes-Consistent Losses for Adaptive Inference in Classification
Abstract
Adaptive inference systems balance accuracy and inference cost by routing inputs to models with varying capabilities. Although jointly training models and routers is empirically more sample-efficient than training them separately, principled losses for joint training remain underexplored. We address this gap in a classification setting by proposing cost-sensitive hinge losses for jointly training classifiers with single-stage or cascade routers. Single-stage routers directly select a model per input, whereas cascade routers sequentially decide at each stage whether to predict with the current model or defer to the next stage, which uses a more accurate but more costly model. For both settings, we prove linear calibration bounds establishing Bayes consistency: minimizing the hinge risk recovers the Bayes-optimal risk under the target cost-sensitive zero-one loss. Moreover, since routers can require more labeled data to train than classifiers, we propose a novel post-training prediction-powered inference (PPI) procedure that uses unlabeled data to refine a trained single-stage router. Experiments on synthetic and real datasets demonstrate that the proposed hinge losses are effective for joint training and that PPI can correct routing errors when labeled data is scarce.
Related
- Beyond Tweedie's Formula: Conditional Score Modeling for Empirical Bayes Inference
- Generalised Robust Bayes for Joint Inference of Model and Contamination
- Adaptive Model-Switch Thermodynamic Integration: A Bayes Factor Estimator
- Variational Bayes for Variable Selection with Adaptive Sharing across Populations and Responses
- Empirical Bayes for compound adaptive experiments
BibTeX
@misc{liu2026bayesconsistent,
title = {{Bayes-Consistent Losses for Adaptive Inference in Classification}},
author = {Shirley Xiaoqi Liu and Florence Regol and Antonios Valkanas and Patrick Rebeschini and Mark Coates},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.39709},
eprint = {2610.39709},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.39709},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Bayes-Consistent Losses for Adaptive Inference in Classification
AU - Shirley Xiaoqi Liu
AU - Florence Regol
AU - Antonios Valkanas
AU - Patrick Rebeschini
AU - Mark Coates
PY - 2026
DA - 2026/10/29
PB - arXiv
DO - 10.48550/arxiv.2610.39709
UR - https://arxiv.org/abs/2610.39709
AB - Adaptive inference systems balance accuracy and inference cost by routing inputs to models with varying capabilities. Although jointly training models and routers is empirically more sample-efficient than training them separately, principled losses for joint training remain underexplored. We address this gap in a classification setting by proposing cost-sensitive hinge losses for jointly training classifiers with single-stage or cascade routers. Single-stage routers directly select a model per input, whereas cascade routers sequentially decide at each stage whether to predict with the current model or defer to the next stage, which uses a more accurate but more costly model. For both settings, we prove linear calibration bounds establishing Bayes consistency: minimizing the hinge risk recovers the Bayes-optimal risk under the target cost-sensitive zero-one loss. Moreover, since routers can require more labeled data to train than classifiers, we propose a novel post-training prediction-powered inference (PPI) procedure that uses unlabeled data to refine a trained single-stage router. Experiments on synthetic and real datasets demonstrate that the proposed hinge losses are effective for joint training and that PPI can correct routing errors when labeled data is scarce.
ER -