NSTL国家科技图书文献中心

1. Magnitude Attention-based Dynamic Pruning NSTL国家科技图书文献中心

Jihye Back | Namhyuk Ahn ... - 《Expert Systems with Application》 - 2025,276(Jun.) - 126957.1～126957.11 - 共11页

摘要：Existing pruning* methods often rely on weight* | Dynamic Pruning* (MAP) method, which applies the* | but also outperform previous pruning* methods on* | importance to identify sparse structures but typically | apply this information statically, without leveraging

关键词： Model compression | Model pruning | Dynamic pruning | Deep learning | Optimization

2. Isomorphic Pruning for Vision Models NSTL国家科技图书文献中心

Gongfan Fang | Xinyin Ma ... - 《Computer Vision - ECCV 2024,Part XXX》 - European Conference on Computer Vision - 2025, - 232～250 - 共19页

摘要：Structured pruning* reduces the computational* | this, we present Isomorphic Pruning, a simple | across different model sizes. Isomorphic Pruning | pruning. Our empirical results on ImageNet-1K | demonstrate that Isomorphic Pruning* surpasses several*

关键词： Network pruning | Vision transformers | CNNs

3. RePaIR: Repaired pruning at initialization resilience NSTL国家科技图书文献中心

Zhao, Haocheng | Guan, Runwei ... - 《Neural Networks》 - 2025,184 - Article 107086～Article 107086 - 共15页

摘要： application of neural network pruning. Unstructured pruning | . Unstructured Pruning* at Initialization (PaI) optimizes the* | iterative pruning* pipeline, but sparse weights increase* | obtaining the best pruning* mask without considering* | , and name it Repaired Pruning* at Initialization*

关键词： Pruning at initialization | Lipschitz | Neural network | Unstructured pruning

4. Effective Layer Pruning Through Similarity Metric Perspective NSTL国家科技图书文献中心

Ian Pons | Bruno Yamamoto ... - 《Pattern Recognition,Part V》 - International Conference on Pattern Recognition - 2025, - 423～438 - 共16页

摘要： demonstrated that pruning* structures from these models is a* | filters. Studies have also been devoted to layer pruning | . However, layer pruning* often hurts the network* | rates. This work introduces an effective layer-pruning | pruning methods. Our method estimates the relative

关键词： Layer pruning | Similarity metric | Efficient deep learning

5. CpdsConv: Continuous Pruning for Depthwise Separable Convolution NSTL国家科技图书文献中心

Yan Ding | Jianbang Xiao ... - 《Tenth Symposium on Novel Optoelectronic Detection Technology and Applications,Part One of Three Parts》 - Symposium on Novel Optoelectronic Detection Technology and Applications - 2025, - 1351111.1～1351111.19 - 共19页

摘要：Current structured pruning* methods typically* | evaluation metrics for pruning, leading to significant | drops in network accuracy post-pruning. This paper | proposes a novel continuous pruning* method for depthwise* | employ fixed thresholds and rank channels based on

关键词： Convolutional neural networks | Structured pruning | Evaluation metrics | Classification

6. LPViT: Low-Power Semi-structured Pruning for Vision Transformers NSTL国家科技图书文献中心

Kaixin Xu | Zhe Wang ... - 《Computer Vision - ECCV 2024,Part LXXI》 - European Conference on Computer Vision - 2025, - 269～287 - 共19页

摘要： introduce a new block-structured pruning* to address the* | . Unlike unstructured pruning* or channel-wise structured* | pruning, block pruning leverages the block-wise | matrix multiplications. To optimize this pruning* scheme* | algorithm to achieve post-training pruning* for ViTs*

关键词： Network pruning | Vision transformers

7. Self-distillation enhanced adaptive pruning of convolutional neural networks NSTL国家科技图书文献中心

Diao, Huabin | Li, Gongyan ... - 《Pattern Recognition》 - 2025,157 - 共11页

摘要： adaptive pruning* algorithm based on self-distillation* | each channel to control channel pruning* and* | integrates the pruning* process into network training* | , enabling pruning* and fine-tuning in a single training* | framework requires only a single overall pruning* rate to*

关键词： Convolutional neural networks | Self-distillation | Adaptive pruning

8. Pruning convolutional neural networks for inductive conformal prediction NSTL国家科技图书文献中心

Zhao X. | Bellotti A.... - 《Neurocomputing》 - 2025,611(Jan.1) - 1.1～1.19 - 共19页

摘要：© 2024Neural network pruning* is a popular* | neural network pruning* in the context of conformal* | nowadays. Therefore, we focus on pruning* CNNs and, in* | particular, filter-level pruning. We first propose a brute | contribution. Furthermore, we improve the global pruning

关键词： Conformal prediction | Convolutional neural network | Filter-level pruning | Neural pruning | Uncertainty estimation

9. Federated adaptive pruning with differential privacy NSTL国家科技图书文献中心

Zhousheng Wang | Jiahe Shen ... - 《Future generations computer systems》 - 2025,169(Aug.) - 107783.1～107783.10 - 共10页

摘要： Adaptive Pruning* (FAP), a lightweight method that* | integrates FL with adaptive pruning* by adjusting explicit* | Federated Learning (FL), as an emerging | distributed machine learning technique, reduces the | computational burden on the central server through

关键词： Federated learning | Lightweight machine learning | Data pruning | Differential privacy

10. Straightforward Layer-Wise Pruning for More Efficient Visual Adaptation NSTL国家科技图书文献中心

Ruizi Han | Jinglei Tang - 《Computer Vision - ECCV 2024,Part LXXII》 - European Conference on Computer Vision - 2025, - 236～252 - 共17页

摘要：. Structural pruning* effectively reduces model redundancy* | ; however, common pruning* methods often lead to an* | pruning structures based on pruning rates and data | Straightforward layer-wise pruning* method, called SLS, for* | pruning PETL-transferred models. By evaluating

摘要：Existing pruning* methods often rely on weight* | Dynamic Pruning* (MAP) method, which applies the* | but also outperform previous pruning* methods on* | importance to identify sparse structures but typically | apply this information statically, without leveraging

关键词： Model compression | Model pruning | Dynamic pruning | Deep learning | Optimization

摘要：Structured pruning* reduces the computational* | this, we present Isomorphic Pruning, a simple | across different model sizes. Isomorphic Pruning | pruning. Our empirical results on ImageNet-1K | demonstrate that Isomorphic Pruning* surpasses several*

关键词： Network pruning | Vision transformers | CNNs

关键词： Pruning at initialization | Lipschitz | Neural network | Unstructured pruning

摘要： demonstrated that pruning* structures from these models is a* | filters. Studies have also been devoted to layer pruning | . However, layer pruning* often hurts the network* | rates. This work introduces an effective layer-pruning | pruning methods. Our method estimates the relative

关键词： Layer pruning | Similarity metric | Efficient deep learning

摘要：Current structured pruning* methods typically* | evaluation metrics for pruning, leading to significant | drops in network accuracy post-pruning. This paper | proposes a novel continuous pruning* method for depthwise* | employ fixed thresholds and rank channels based on

关键词： Convolutional neural networks | Structured pruning | Evaluation metrics | Classification

摘要： introduce a new block-structured pruning* to address the* | . Unlike unstructured pruning* or channel-wise structured* | pruning, block pruning leverages the block-wise | matrix multiplications. To optimize this pruning* scheme* | algorithm to achieve post-training pruning* for ViTs*

关键词： Network pruning | Vision transformers

摘要： adaptive pruning* algorithm based on self-distillation* | each channel to control channel pruning* and* | integrates the pruning* process into network training* | , enabling pruning* and fine-tuning in a single training* | framework requires only a single overall pruning* rate to*

关键词： Convolutional neural networks | Self-distillation | Adaptive pruning

摘要：© 2024Neural network pruning* is a popular* | neural network pruning* in the context of conformal* | nowadays. Therefore, we focus on pruning* CNNs and, in* | particular, filter-level pruning. We first propose a brute | contribution. Furthermore, we improve the global pruning

关键词： Conformal prediction | Convolutional neural network | Filter-level pruning | Neural pruning | Uncertainty estimation

摘要： Adaptive Pruning* (FAP), a lightweight method that* | integrates FL with adaptive pruning* by adjusting explicit* | Federated Learning (FL), as an emerging | distributed machine learning technique, reduces the | computational burden on the central server through

关键词： Federated learning | Lightweight machine learning | Data pruning | Differential privacy

摘要：. Structural pruning* effectively reduces model redundancy* | ; however, common pruning* methods often lead to an* | pruning structures based on pruning rates and data | Straightforward layer-wise pruning* method, called SLS, for* | pruning PETL-transferred models. By evaluating

关键词： Parameter-efficient transfer learning | Network pruning | T-SNE

4008-161-200 800-990-8900

国家科技图书文献中心

摘要：Existing pruning methods often rely on weight | Dynamic Pruning (MAP) method, which applies the | but also outperform previous pruning methods on | importance to identify sparse structures but typically | apply this information statically, without leveraging

关键词： Model compression | Model pruning | Dynamic pruning | Deep learning | Optimization

摘要：Structured pruning reduces the computational | this, we present Isomorphic Pruning, a simple | across different model sizes. Isomorphic Pruning | pruning. Our empirical results on ImageNet-1K | demonstrate that Isomorphic Pruning surpasses several

关键词： Network pruning | Vision transformers | CNNs

摘要： application of neural network pruning. Unstructured pruning | . Unstructured Pruning at Initialization (PaI) optimizes the | iterative pruning pipeline, but sparse weights increase | obtaining the best pruning mask without considering | , and name it Repaired Pruning at Initialization

关键词： Pruning at initialization | Lipschitz | Neural network | Unstructured pruning

摘要： demonstrated that pruning structures from these models is a | filters. Studies have also been devoted to layer pruning | . However, layer pruning often hurts the network | rates. This work introduces an effective layer-pruning | pruning methods. Our method estimates the relative

关键词： Layer pruning | Similarity metric | Efficient deep learning

摘要：Current structured pruning methods typically | evaluation metrics for pruning, leading to significant | drops in network accuracy post-pruning. This paper | proposes a novel continuous pruning method for depthwise | employ fixed thresholds and rank channels based on

关键词： Convolutional neural networks | Structured pruning | Evaluation metrics | Classification

摘要： introduce a new block-structured pruning to address the | . Unlike unstructured pruning or channel-wise structured | pruning, block pruning leverages the block-wise | matrix multiplications. To optimize this pruning scheme | algorithm to achieve post-training pruning for ViTs

关键词： Network pruning | Vision transformers

摘要： adaptive pruning algorithm based on self-distillation | each channel to control channel pruning and | integrates the pruning process into network training | , enabling pruning and fine-tuning in a single training | framework requires only a single overall pruning rate to

关键词： Convolutional neural networks | Self-distillation | Adaptive pruning

摘要：© 2024Neural network pruning is a popular | neural network pruning in the context of conformal | nowadays. Therefore, we focus on pruning CNNs and, in | particular, filter-level pruning. We first propose a brute | contribution. Furthermore, we improve the global pruning

关键词： Conformal prediction | Convolutional neural network | Filter-level pruning | Neural pruning | Uncertainty estimation

摘要： Adaptive Pruning (FAP), a lightweight method that | integrates FL with adaptive pruning by adjusting explicit | Federated Learning (FL), as an emerging | distributed machine learning technique, reduces the | computational burden on the central server through

关键词： Federated learning | Lightweight machine learning | Data pruning | Differential privacy

摘要：. Structural pruning effectively reduces model redundancy | ; however, common pruning methods often lead to an | pruning structures based on pruning rates and data | Straightforward layer-wise pruning method, called SLS, for | pruning PETL-transferred models. By evaluating

关键词： Parameter-efficient transfer learning | Network pruning | T-SNE

4008-161-200 800-990-8900

国家科技图书文献中心

摘要：Existing pruning* methods often rely on weight* | Dynamic Pruning* (MAP) method, which applies the* | but also outperform previous pruning* methods on* | importance to identify sparse structures but typically | apply this information statically, without leveraging

摘要：Structured pruning* reduces the computational* | this, we present Isomorphic Pruning, a simple | across different model sizes. Isomorphic Pruning | pruning. Our empirical results on ImageNet-1K | demonstrate that Isomorphic Pruning* surpasses several*

摘要： demonstrated that pruning* structures from these models is a* | filters. Studies have also been devoted to layer pruning | . However, layer pruning* often hurts the network* | rates. This work introduces an effective layer-pruning | pruning methods. Our method estimates the relative

摘要：Current structured pruning* methods typically* | evaluation metrics for pruning, leading to significant | drops in network accuracy post-pruning. This paper | proposes a novel continuous pruning* method for depthwise* | employ fixed thresholds and rank channels based on

摘要： introduce a new block-structured pruning* to address the* | . Unlike unstructured pruning* or channel-wise structured* | pruning, block pruning leverages the block-wise | matrix multiplications. To optimize this pruning* scheme* | algorithm to achieve post-training pruning* for ViTs*

摘要： adaptive pruning* algorithm based on self-distillation* | each channel to control channel pruning* and* | integrates the pruning* process into network training* | , enabling pruning* and fine-tuning in a single training* | framework requires only a single overall pruning* rate to*

摘要：© 2024Neural network pruning* is a popular* | neural network pruning* in the context of conformal* | nowadays. Therefore, we focus on pruning* CNNs and, in* | particular, filter-level pruning. We first propose a brute | contribution. Furthermore, we improve the global pruning

摘要： Adaptive Pruning* (FAP), a lightweight method that* | integrates FL with adaptive pruning* by adjusting explicit* | Federated Learning (FL), as an emerging | distributed machine learning technique, reduces the | computational burden on the central server through

摘要：. Structural pruning* effectively reduces model redundancy* | ; however, common pruning* methods often lead to an* | pruning structures based on pruning rates and data | Straightforward layer-wise pruning* method, called SLS, for* | pruning PETL-transferred models. By evaluating