aiqbench analysis
This research introduces a correlation-aware structured pruning technique that explicitly models dependencies between model units to avoid the performance degradation caused by ignoring activation correlations. By formulating the optimization as a binary quadratic program solved via a greedy interaction algorithm, the method achieves competitive accuracy-efficiency trade-offs across mainstream large language models. The work advances the field of efficient AI inference by refining how sparse architectures are constructed without sacrificing predictive capability.
Correlation-Aware Structured Pruning for Large Language Models
arXiv:2609.22131v1 Announce Type: new Abstract: Structured pruning is a promising approach for reducing the substantial inference costs of Large Language Models (LLMs) while maintaining hardware efficiency. Many existing methods assess the importance of prunable units (e.g., channels or heads) in isolation, implicitly assuming that pruning errors are additive. This independence assumption is often invalidated by the non-orthogonality of model weights and strong correlations between unit activations, potentially leading to performance degradation. To address this, we propose a Correlation-Aware Structured Pruning method. We formulate the pruning objective as a cardinality-constrained binary quadratic program that explicitly models cross-unit dependencies in the reconstruction error. Since this binary quadratic program is NP-hard and difficult to solve exactly, we develop a greedy interaction algorithm based on dependency-aware marginal costs to optimize unit selection. Furthermore, we incorporate a gradient-based strategy to achieve adaptive layer-wise sparsity allocation across the entire model. Extensive experiments on mainstream LLMs demonstrate that incorporating correlation information yields competitive accuracy-efficiency trade-offs compared to representative structured pruning baselines.
Abstract:Structured pruning is a promising approach for reducing the substantial inference costs of Large Language Models (LLMs) while maintaining hardware efficiency. Many existing methods assess the importance of prunable units (e.g., channels or heads) in isolation, implicitly assuming that pruning errors are additive. This independence assumption is often invalidated by the non-orthogonality of model weights and strong correlations between unit activations, potentially leading to performance degradation. To address this, we propose a Correlation-Aware Structured Pruning method. We formulate the pruning objective as a cardinality-constrained binary quadratic program that explicitly models cross-unit dependencies in the reconstruction error. Since this binary quadratic program is NP-hard and difficult to solve exactly, we develop a greedy interaction algorithm based on dependency-aware marginal costs to optimize unit selection. Furthermore, we incorporate a gradient-based strategy to achieve adaptive layer-wise sparsity allocation across the entire model. Extensive experiments on mainstream LLMs demonstrate that incorporating correlation information yields competitive accuracy-efficiency trade-offs compared to representative structured pruning baselines.
Computation and Language (cs.CL); Machine Learning (cs.LG)
arXiv:2609.22131 [cs.CL]
(or arXiv:2609.22131v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.22131 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 255 words
Related coverage
PIRAMID: Progress and Plans
LessWrong · Jul 27, 2026
[AINews] not much happened today
Latent Space · Sep 10, 2026
Partition-Aware Unlearning for Removing Spurious Correlations in Large Vision-Language Models
arXiv · Sep 1, 2026
Heterogeneous Structured Pruning Under Resource Constraints for Hardware-Efficient Large Language Models
arXiv · Oct 14, 2026