Risk-Bounded Distribution Reconstruction: Stable Statistic Calibration for Long-Tailed Recognition

Guanliang Liu, Wenchao Chen, Long Tian, Xuefei Cao, Hongwei Liu, Bo Chen
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:75774-75803, 2026.

Abstract

Long-tailed recognition suffers from extreme class imbalance, where scarce tail data leads to biased and fragile feature distributions that exacerbate confusion with semantically or visually similar classes. Prior feature-space reconstruction methods transfer head-class structure or train conditional generators to synthesize tail features, yet the resulting statistical updates are often heuristic and can degrade multi-class separability when tail estimates are unreliable. Given this issue, we propose Risk-Bounded Distribution Reconstruction (RBDR), an offline statistic calibration framework for the two-stage long-tailed pipeline, grounded in an analysis of rival-induced discriminative directions. RBDR performs (i) risk-aware mean calibration by softly projecting any candidate update onto a supportive set such that a surrogate discriminative-risk upper bound does not increase, and (ii) covariance control by shrinking dispersion in a rival subspace while preserving diversity in orthogonal components. In the frozen-feature reconstruction setting, these calibration components convert heuristic reconstruction cues into controllable distributional updates, improving both performance and stability across multiple long-tailed benchmarks.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-liu26t, title = {Risk-Bounded Distribution Reconstruction: Stable Statistic Calibration for Long-Tailed Recognition}, author = {Liu, Guanliang and Chen, Wenchao and Tian, Long and Cao, Xuefei and Liu, Hongwei and Chen, Bo}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {75774--75803}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/liu26t/liu26t.pdf}, url = {https://proceedings.mlr.press/v306/liu26t.html}, abstract = {Long-tailed recognition suffers from extreme class imbalance, where scarce tail data leads to biased and fragile feature distributions that exacerbate confusion with semantically or visually similar classes. Prior feature-space reconstruction methods transfer head-class structure or train conditional generators to synthesize tail features, yet the resulting statistical updates are often heuristic and can degrade multi-class separability when tail estimates are unreliable. Given this issue, we propose Risk-Bounded Distribution Reconstruction (RBDR), an offline statistic calibration framework for the two-stage long-tailed pipeline, grounded in an analysis of rival-induced discriminative directions. RBDR performs (i) risk-aware mean calibration by softly projecting any candidate update onto a supportive set such that a surrogate discriminative-risk upper bound does not increase, and (ii) covariance control by shrinking dispersion in a rival subspace while preserving diversity in orthogonal components. In the frozen-feature reconstruction setting, these calibration components convert heuristic reconstruction cues into controllable distributional updates, improving both performance and stability across multiple long-tailed benchmarks.} }
Endnote
%0 Conference Paper %T Risk-Bounded Distribution Reconstruction: Stable Statistic Calibration for Long-Tailed Recognition %A Guanliang Liu %A Wenchao Chen %A Long Tian %A Xuefei Cao %A Hongwei Liu %A Bo Chen %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-liu26t %I PMLR %P 75774--75803 %U https://proceedings.mlr.press/v306/liu26t.html %V 306 %X Long-tailed recognition suffers from extreme class imbalance, where scarce tail data leads to biased and fragile feature distributions that exacerbate confusion with semantically or visually similar classes. Prior feature-space reconstruction methods transfer head-class structure or train conditional generators to synthesize tail features, yet the resulting statistical updates are often heuristic and can degrade multi-class separability when tail estimates are unreliable. Given this issue, we propose Risk-Bounded Distribution Reconstruction (RBDR), an offline statistic calibration framework for the two-stage long-tailed pipeline, grounded in an analysis of rival-induced discriminative directions. RBDR performs (i) risk-aware mean calibration by softly projecting any candidate update onto a supportive set such that a surrogate discriminative-risk upper bound does not increase, and (ii) covariance control by shrinking dispersion in a rival subspace while preserving diversity in orthogonal components. In the frozen-feature reconstruction setting, these calibration components convert heuristic reconstruction cues into controllable distributional updates, improving both performance and stability across multiple long-tailed benchmarks.
APA
Liu, G., Chen, W., Tian, L., Cao, X., Liu, H. & Chen, B.. (2026). Risk-Bounded Distribution Reconstruction: Stable Statistic Calibration for Long-Tailed Recognition. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:75774-75803 Available from https://proceedings.mlr.press/v306/liu26t.html.

Related Material