A Robust Optimization Guided Pruning Framework for Vision and Large Language Models

Gabriel Afriat, Hussein Hazimeh, Dimitris Paparas, Rahul Mazumder
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:690-714, 2026.

Abstract

Pruning is a common approach to reduce the memory footprint and inference cost of large vision and language models. As these architectures continue to scale, one-shot pruning methods - i.e. approaches that prune the network without any retraining - have become increasingly attractive. Many popular one-shot pruning methods (e.g., WoodFisher, CAP, SparseGPT, and ALPS) typically optimize a quadratic objective under sparsity constraints. However, in practice, this objective is affected by multiple sources of uncertainty, including noise in the calibration data and variability introduced by algorithmic updates. To address these issues, we introduce RobOP, a robust optimization framework that explicitly accounts for such uncertainties. RobOP is modular and flexible, and can be applied with any existing pruning method through simple modifications motivated by our theoretical framework. We demonstrate that by taking into account uncertainty, RobOP offers improvements over prior pruning approaches. Our framework applies tractably across a range of stylized uncertainty sets, enabling robust one-shot pruning at scale. Our code is available at https://github.com/mazumder-lab/RobOP.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-afriat26a, title = {A Robust Optimization Guided Pruning Framework for Vision and Large Language Models}, author = {Afriat, Gabriel and Hazimeh, Hussein and Paparas, Dimitris and Mazumder, Rahul}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {690--714}, 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/afriat26a/afriat26a.pdf}, url = {https://proceedings.mlr.press/v306/afriat26a.html}, abstract = {Pruning is a common approach to reduce the memory footprint and inference cost of large vision and language models. As these architectures continue to scale, one-shot pruning methods - i.e. approaches that prune the network without any retraining - have become increasingly attractive. Many popular one-shot pruning methods (e.g., WoodFisher, CAP, SparseGPT, and ALPS) typically optimize a quadratic objective under sparsity constraints. However, in practice, this objective is affected by multiple sources of uncertainty, including noise in the calibration data and variability introduced by algorithmic updates. To address these issues, we introduce RobOP, a robust optimization framework that explicitly accounts for such uncertainties. RobOP is modular and flexible, and can be applied with any existing pruning method through simple modifications motivated by our theoretical framework. We demonstrate that by taking into account uncertainty, RobOP offers improvements over prior pruning approaches. Our framework applies tractably across a range of stylized uncertainty sets, enabling robust one-shot pruning at scale. Our code is available at https://github.com/mazumder-lab/RobOP.} }
Endnote
%0 Conference Paper %T A Robust Optimization Guided Pruning Framework for Vision and Large Language Models %A Gabriel Afriat %A Hussein Hazimeh %A Dimitris Paparas %A Rahul Mazumder %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-afriat26a %I PMLR %P 690--714 %U https://proceedings.mlr.press/v306/afriat26a.html %V 306 %X Pruning is a common approach to reduce the memory footprint and inference cost of large vision and language models. As these architectures continue to scale, one-shot pruning methods - i.e. approaches that prune the network without any retraining - have become increasingly attractive. Many popular one-shot pruning methods (e.g., WoodFisher, CAP, SparseGPT, and ALPS) typically optimize a quadratic objective under sparsity constraints. However, in practice, this objective is affected by multiple sources of uncertainty, including noise in the calibration data and variability introduced by algorithmic updates. To address these issues, we introduce RobOP, a robust optimization framework that explicitly accounts for such uncertainties. RobOP is modular and flexible, and can be applied with any existing pruning method through simple modifications motivated by our theoretical framework. We demonstrate that by taking into account uncertainty, RobOP offers improvements over prior pruning approaches. Our framework applies tractably across a range of stylized uncertainty sets, enabling robust one-shot pruning at scale. Our code is available at https://github.com/mazumder-lab/RobOP.
APA
Afriat, G., Hazimeh, H., Paparas, D. & Mazumder, R.. (2026). A Robust Optimization Guided Pruning Framework for Vision and Large Language Models. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:690-714 Available from https://proceedings.mlr.press/v306/afriat26a.html.

Related Material