Calibration effects of pseudo-label noise in semi-supervised training

Determine how error-correlated noise introduced by pseudo-labels in semi-supervised training affects calibration for brain-tumour segmentation models trained under the BraTS-GoAT data constraints.

Background

The paper identifies semi-supervised training as a possible extension but notes that the BraTS-GoAT challenge permits no external data. Although pseudo-labels could enable semi-supervised learning using available data, they may introduce noise correlated with model errors. The authors explicitly leave unresolved how this noise would affect the calibration of the resulting segmentation models.

References

Semi-supervised training is a natural extension, but GoAT permits no external data, and pseudo-labels inject error-correlated noise whose effect on calibration is itself an open question.

— Reliability analysis for BraTS-GoAT segmentation: a controlled robustness study of deep-ensemble uncertainty  (2608.13223 - Shet et al., 13 Aug 2026) in Discussion, final paragraph of the limitations section

Future work should investigate these interactions and test whether CutMix offers a general mechanism for enhancing reliability and robustness across tasks and modalities by extending evaluations to other domains, such as medical imaging or remote sensing.

— The Impact of CutMix on Reliability and Robustness in Semantic Segmentation  (2608.18715 - Landgraf et al., 19 Aug 2026) in Section Conclusion