Dynamic Multimodal Evaluation via Knowledge-Enhanced Benchmark Evolution

Junzhe Zhang, Huixuan Zhang, Xiaojun Wan
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:160695-160715, 2026.

Abstract

The rapid development of multimodal large language models (MLLMs) has created an urgent demand for more reliable and robust evaluation protocols, however, existing static benchmarks are prone to data contamination and performance saturation, which can result in inflated or misleading evaluation results. To address these limitations, we first introduce a graph formulation to represent both static and dynamic visual question answering (VQA) samples. Building upon this formulation, we propose Knowledge-Enhanced Benchmark Evolution (KBE), a dynamic multimodal evaluation framework that first analyzes the original static benchmark, then expands it by integrating multimodal knowledge, transforming the static benchmark into a controllable, dynamic evolving version. Crucially, KBE can both reconstruct questions by Re-selecting visual information in the original image and expand existing questions with external textual knowledge. By explicitly controlling the degree of question exploration, KBE enables difficulty-controllable evaluation across a wide range of model capabilities. Extensive experimental results demonstrate that KBE effectively mitigates data contamination and benchmark saturation, while providing a more comprehensive and flexible assessment of MLLM performance.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-zhang26it, title = {Dynamic Multimodal Evaluation via Knowledge-Enhanced Benchmark Evolution}, author = {Zhang, Junzhe and Zhang, Huixuan and Wan, Xiaojun}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {160695--160715}, 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/zhang26it/zhang26it.pdf}, url = {https://proceedings.mlr.press/v306/zhang26it.html}, abstract = {The rapid development of multimodal large language models (MLLMs) has created an urgent demand for more reliable and robust evaluation protocols, however, existing static benchmarks are prone to data contamination and performance saturation, which can result in inflated or misleading evaluation results. To address these limitations, we first introduce a graph formulation to represent both static and dynamic visual question answering (VQA) samples. Building upon this formulation, we propose Knowledge-Enhanced Benchmark Evolution (KBE), a dynamic multimodal evaluation framework that first analyzes the original static benchmark, then expands it by integrating multimodal knowledge, transforming the static benchmark into a controllable, dynamic evolving version. Crucially, KBE can both reconstruct questions by Re-selecting visual information in the original image and expand existing questions with external textual knowledge. By explicitly controlling the degree of question exploration, KBE enables difficulty-controllable evaluation across a wide range of model capabilities. Extensive experimental results demonstrate that KBE effectively mitigates data contamination and benchmark saturation, while providing a more comprehensive and flexible assessment of MLLM performance.} }
Endnote
%0 Conference Paper %T Dynamic Multimodal Evaluation via Knowledge-Enhanced Benchmark Evolution %A Junzhe Zhang %A Huixuan Zhang %A Xiaojun Wan %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-zhang26it %I PMLR %P 160695--160715 %U https://proceedings.mlr.press/v306/zhang26it.html %V 306 %X The rapid development of multimodal large language models (MLLMs) has created an urgent demand for more reliable and robust evaluation protocols, however, existing static benchmarks are prone to data contamination and performance saturation, which can result in inflated or misleading evaluation results. To address these limitations, we first introduce a graph formulation to represent both static and dynamic visual question answering (VQA) samples. Building upon this formulation, we propose Knowledge-Enhanced Benchmark Evolution (KBE), a dynamic multimodal evaluation framework that first analyzes the original static benchmark, then expands it by integrating multimodal knowledge, transforming the static benchmark into a controllable, dynamic evolving version. Crucially, KBE can both reconstruct questions by Re-selecting visual information in the original image and expand existing questions with external textual knowledge. By explicitly controlling the degree of question exploration, KBE enables difficulty-controllable evaluation across a wide range of model capabilities. Extensive experimental results demonstrate that KBE effectively mitigates data contamination and benchmark saturation, while providing a more comprehensive and flexible assessment of MLLM performance.
APA
Zhang, J., Zhang, H. & Wan, X.. (2026). Dynamic Multimodal Evaluation via Knowledge-Enhanced Benchmark Evolution. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:160695-160715 Available from https://proceedings.mlr.press/v306/zhang26it.html.

Related Material