首页/论文/1901.11391

arXiv 1901.113912019-02-27cs.CV

Partition Pruning: Parallelization-Aware Pruning for Deep Neural Networks

University of California,Irvine（加州大学欧文分校）

Sina Shahhosseini, Ahmad Albaqsami, Masoomeh Jasemi, Nader Bagherzadeh

英文摘要：

Parameters of recent neural networks require a huge amount of memory. These parameters are used by neural networks to perform machine learning tasks when processing inputs. To speed up inference, we develop Partition Pruning, an innovative scheme to reduce the parameters used while taking into consideration parallelization. We evaluated the performance and energy consumption of parallel inference of partitioned models, which showed a 7.72x speed up of performance and a 2.73x reduction in the energy used for computing pruned layers of TinyVGG16 in comparison to running the unpruned model on a single accelerator. In addition, our method showed a limited reduction some numbers in accuracy while partitioning fully connected layers.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。

阅读原文

阅读 PDF ↗

arXiv 页面 ↗HTML 原文 ↗

本文内容

英文摘要相关论文

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。

继续探索

相关论文

SGLP：一种用于压缩大型深度模型的相似度引导快速层分区剪枝方法

SGLP: A Similarity Guided Fast Layer Partition Pruning for Compressing Large Deep Models

Institute of Computing Technology, Chinese Academy of Sciences（中国科学院计算技术研究所）；Zhejiang University of Technology（浙江工业大学）；Nanyang Technological University（南洋理工大学）；Zhejiang University（浙江大学）；Southwest University（西南大学）；The City College of New York（纽约城市学院）；Shenzhen University of Advanced Technology（深圳大学）

针对现有层剪枝方法忽略层间依赖易损失预训练模型核心特征的问题，提出SGLP相似度引导层分区剪枝框架，结合CKA相似度计算、Fisher最优分割与GradNorm免微调评估，在图像分类和LLM任务上实现了更优的压缩精度与效率。

2025-11-14cs.CVcs.LG

Novel Adaptive Binary Search Strategy-First Hybrid Pyramid- and Clustering-Based CNN Filter Pruning Method without Parameters Setting

National Taiwan University of Science and Technology（国立台湾科技大学）

Pruning redundant filters in CNN models has received growing attention. In this paper, we propose an adaptive binary search-first hybrid pyramid- and clustering-based (ABSHPC-based) method for pruning filters automatically. In our method, for each convolutional layer, initially a hybrid pyramid data structure is constructed to store the hierarchical information of each filter. Given a tolerant accuracy loss, without parameters setting, we begin from the last convolutional layer to the first layer; for each considered layer with less or equal pruning rate relative to its previous layer, our ABS…

2021-04-30cs.CV

通过视图图剪枝实现鲁棒的全局运动恢复结构（Structure-from-Motion, SfM）

Robust Global Structure-from-Motion via View Graph Pruning

Hangzhou Dianzi University（杭州电子科技大学）；State Key Lab of CAD&CG, Zhejiang University（浙江大学CAD&CG国家重点实验室）

本研究针对全局运动恢复结构（SfM）易受错误边影响的问题，提出子图引导的视图图剪枝框架，经多类数据集实验验证可提升其鲁棒性并优化新视图合成质量。

2026-08-25cs.CV

FastVID: 动态密度修剪用于快速视频大语言模型

FastVID: Dynamic Density Pruning for Fast Video Large Language Models

School of Software, Tsinghua University（清华大学软件学院）；BNRist, Tsinghua University（清华大学BNRist）；GRG Banking Equipment Co., Ltd.（GRG银行设备有限公司）

FastVID通过动态密度修剪技术，显著降低视频大语言模型的计算开销，同时保持高准确性，实现高效的视频处理性能。

2025-12-14cs.CV

HAP：基于跨模态对齐的头部自适应视觉Token剪枝

HAP: Head-Adaptive Visual Token Pruning via Cross-Modal Alignment

Shanghai Jiao Tong University（上海交通大学）；University of Edinburgh（爱丁堡大学）

该研究针对视觉-语言模型预填充成本过高的问题，提出基于PAQ指标的头部自适应视觉Token剪枝方法，在18个基准上实现最优权衡，在LLaVA-1.5-7B上仅保留5.6% Token即可维持99.1%的原始性能，优于基线AutoPrune。

2026-08-26cs.CV

通过信息最大化进行数据剪枝

Data Pruning by Information Maximization

The University of Hong Kong（香港大学）；The Chinese University of Hong Kong（香港中文大学）

提出InfoMax数据剪枝方法，通过重要性得分和成对相似度将核心集选择形式化为DQP任务，并引入梯度求解器实现大规模扩展，在多种任务中表现卓越。

2025-08-14cs.AIcs.CV

更多相关论文收起更多论文

GSTEP：面向高效视频大语言模型的全局时空密度驱动视觉令牌剪枝

GSTEP: Global Spatio-Temporal Density-Driven Visual Token Pruning for Efficient Video Large Language Models

本文提出即插即用的GSTEP剪枝框架，解决现有视频令牌剪枝方法的局部剪枝缺陷，在多个VideoLLMs上实现良好的准确率-效率权衡，在LLaVA-OneVision-7B上剪去75%视觉令牌，保留100.2%原始性能并获1.17倍端到端加速。

2026-08-05cs.CLcs.CV

Zero-TPrune：通过利用预训练 Transformer 中的注意力图实现零样本 Token 剪枝

Zero-TPrune: Zero-Shot Token Pruning through Leveraging of the Attention Graph in Pre-Trained Transformers

Princeton University（普林斯顿大学）

本文提出零样本 token 剪枝方法 Zero-TPrune，利用预训练 Transformer 注意力图和 WPR 算法综合评估 token 重要性与相似性，无需微调即可显著降低视觉 Transformer 的 FLOPs 并提升吞吐量。

2024-04-07cs.AIcs.CVcs.LG

通过区域、令牌和指令引导的重要性进行高分辨率大视觉-语言模型的金字塔令牌修剪

Pyramid Token Pruning for High-Resolution Large Vision-Language Models via Region, Token, and Instruction-Guided Importance

Shanghai Key Laboratory of Intelligent Information Processing（上海智能信息处理重点实验室）；College of Computer Science and Artificial Intelligence（计算机科学与人工智能学院）

通过区域、令牌和指令引导的重要性进行高分辨率大视觉-语言模型的金字塔令牌修剪，有效降低计算和内存开销，同时保持性能

2025-09-29cs.CV

视差：面向异构边缘系统的运行时并行化以优化操作符回退

Parallax: Runtime Parallelization for Operator Fallbacks in Heterogeneous Edge Systems

University of Southampton（南安普顿大学）；UCL AI Centre（伦敦大学学院人工智能中心）；University College London（伦敦大学学院）

Parallax通过并行化和内存优化，提升移动设备上动态神经网络推断的效率与响应性。

2025-12-12cs.AIcs.CVcs.DC

SplatQ：基于二次选择的感知交互三维高斯溅射剪枝算法

SplatQ: Interaction-Aware Pruning of 3D Gaussian Splatting via Quadratic Selection

SplatQ是一种协调区域预算分配与联合高斯选择的组合剪枝算法，通过二次优化提升3DGS的重建质量均衡性与渲染速度，在5个基准上优于现有最优方法。

2026-10-12cs.CVcs.GR

从修补到剪枝：视觉语言模型中的视觉计算

From Patching to Pruning Visual Computation in Vision Language Models

Northeastern University（东北大学）；EmbodyX Inc.（EmbodyX公司）；Zhejiang University（浙江大学）；Stevens Institute of Technology（史蒂文斯理工学院）

P2P框架通过将激活修补转化为推理时计算旁路，在保留序列结构的同时剪枝视觉计算，实现3%容差下94%精度和55%FLOPs减少，并揭示VLM视觉处理在深度上的非均匀分布。

2026-10-05cs.CVcs.LG

Parallel Training of GRU Networks with a Multi-Grid Solver for Long Sequences

Korea Aerospace University（韩国航空大学）；Sandia National Laboratories（桑迪亚国家实验室）

Parallelizing Gated Recurrent Unit (GRU) networks is a challenging task, as the training procedure of GRU is inherently sequential. Prior efforts to parallelize GRU have largely focused on conventional parallelization strategies such as data-parallel and model-parallel training algorithms. However, when the given sequences are very long, existing approaches are still inevitably performance limited in terms of training time. In this paper, we present a novel parallel training scheme (called parallel-in-time) for GRU based on a multigrid reduction in time (MGRIT) solver. MGRIT partitions a seque…

2022-03-07cs.CVcs.DCcs.LG

Ensemble learning of diffractive optical networks

University of California, Los Angeles（加利福尼亚大学洛杉矶分校）；California NanoSystems Institute (CNSI)（加州纳米系统研究所）

A plethora of research advances have emerged in the fields of optics and photonics that benefit from harnessing the power of machine learning. Specifically, there has been a revival of interest in optical computing hardware, due to its potential advantages for machine learning tasks in terms of parallelization, power efficiency and computation speed. Diffractive Deep Neural Networks (D2NNs) form such an optical computing framework, which benefits from deep learning-based design of successive diffractive layers to all-optically process information as the input light diffracts through these pass…

2020-09-15cs.CVcs.LGcs.NE