Injecting Distributional Awareness into MLLMs via Reinforcement Learning for Deep Imbalanced Regression

Yao Du, Shanshan Song, Xiaomeng Li
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:26602-26623, 2026.

Abstract

Multimodal large language models (MLLMs) struggle with numerical regression under longtailed target distributions. Token-level supervised fine-tuning (SFT) and point-wise regression rewards bias learning toward high-density regions, leading to regression-to-the-mean behavior and poor tail performance. We identify the lack of cross-sample relational supervision as a key limitation of existing MLLM training paradigms. To address it, we propose a distribution-aware reinforcement learning framework based on Group Relative Policy Optimization, which introduces batch-level comparison-based supervision via the Concordance Correlation Coefficient-based reward to align predicted and ground-truth distributions in terms of correlation, scale, and mean. The framework is plug-and-play, requiring no architectural modification. Experiments on a unified suite of long-tailed regression benchmarks show consistent improvements over SFT and existing MLLM regression methods, with particularly strong gains in medium- and few-shot regimes.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-du26j, title = {Injecting Distributional Awareness into {MLLM}s via Reinforcement Learning for Deep Imbalanced Regression}, author = {Du, Yao and Song, Shanshan and Li, Xiaomeng}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {26602--26623}, 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/du26j/du26j.pdf}, url = {https://proceedings.mlr.press/v306/du26j.html}, abstract = {Multimodal large language models (MLLMs) struggle with numerical regression under longtailed target distributions. Token-level supervised fine-tuning (SFT) and point-wise regression rewards bias learning toward high-density regions, leading to regression-to-the-mean behavior and poor tail performance. We identify the lack of cross-sample relational supervision as a key limitation of existing MLLM training paradigms. To address it, we propose a distribution-aware reinforcement learning framework based on Group Relative Policy Optimization, which introduces batch-level comparison-based supervision via the Concordance Correlation Coefficient-based reward to align predicted and ground-truth distributions in terms of correlation, scale, and mean. The framework is plug-and-play, requiring no architectural modification. Experiments on a unified suite of long-tailed regression benchmarks show consistent improvements over SFT and existing MLLM regression methods, with particularly strong gains in medium- and few-shot regimes.} }
Endnote
%0 Conference Paper %T Injecting Distributional Awareness into MLLMs via Reinforcement Learning for Deep Imbalanced Regression %A Yao Du %A Shanshan Song %A Xiaomeng Li %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-du26j %I PMLR %P 26602--26623 %U https://proceedings.mlr.press/v306/du26j.html %V 306 %X Multimodal large language models (MLLMs) struggle with numerical regression under longtailed target distributions. Token-level supervised fine-tuning (SFT) and point-wise regression rewards bias learning toward high-density regions, leading to regression-to-the-mean behavior and poor tail performance. We identify the lack of cross-sample relational supervision as a key limitation of existing MLLM training paradigms. To address it, we propose a distribution-aware reinforcement learning framework based on Group Relative Policy Optimization, which introduces batch-level comparison-based supervision via the Concordance Correlation Coefficient-based reward to align predicted and ground-truth distributions in terms of correlation, scale, and mean. The framework is plug-and-play, requiring no architectural modification. Experiments on a unified suite of long-tailed regression benchmarks show consistent improvements over SFT and existing MLLM regression methods, with particularly strong gains in medium- and few-shot regimes.
APA
Du, Y., Song, S. & Li, X.. (2026). Injecting Distributional Awareness into MLLMs via Reinforcement Learning for Deep Imbalanced Regression. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:26602-26623 Available from https://proceedings.mlr.press/v306/du26j.html.

Related Material