Pruning-Filter-in-Filter:筛选器中的修剪筛选器(NeurIPS2020)

上传者: 42128393 | 上传时间: 2022-07-08 16:12:34 | 文件大小: 3.68MB | 文件类型: ZIP
过滤器中的修剪过滤器 介绍 这是NeurIPS 2020论文“”的PyTorch实施。 在本文中: 我们提出了一种新的修剪模式,称为条带化修剪(SP),可以将其视为过滤修剪(FP)的一般情况。 SP将过滤器$ F \ in \ mathbb {R} ^ {C \ timesK \ times}}视为$ K \ timesK $条带(即,$ 1 \ times $$过滤器$ \ in \ mathbb {R} ^ c $),并以条带为单位而不是整个过滤器执行修剪。 与现有方法相比,SP具有比传统FP更好的粒度,同时比Weight-Pruning更加硬件友好,并且与Group-wise Pruning相比保持了过滤器之间的独立性,从而在CIFAR-10和ImageNet上实现了最先进的修剪率。 更令人振奋的是,通过应用SP,我们发现过滤器的另一个重要特性与重量无关:形状。 从随机初始化的R

文件下载

资源详情

[{"title":"( 30 个子文件 3.68MB ) Pruning-Filter-in-Filter:筛选器中的修剪筛选器(NeurIPS2020)","children":[{"title":"Pruning-Filter-in-Filter-master","children":[{"title":"models","children":[{"title":"resnet56.py <span style='color:#111;'> 2.83KB </span>","children":null,"spread":false},{"title":"__init__.py <span style='color:#111;'> 65B </span>","children":null,"spread":false},{"title":"stripe.py <span style='color:#111;'> 4.71KB </span>","children":null,"spread":false},{"title":"vgg.py <span style='color:#111;'> 2.04KB </span>","children":null,"spread":false}],"spread":true},{"title":"main.py <span style='color:#111;'> 5.72KB </span>","children":null,"spread":false},{"title":"BrokenNet_filter.png <span style='color:#111;'> 153.71KB </span>","children":null,"spread":false},{"title":"xxx-wise.pptx <span style='color:#111;'> 120.99KB </span>","children":null,"spread":false},{"title":"filter skeleton.pptx <span style='color:#111;'> 62.37KB </span>","children":null,"spread":false},{"title":"flops.py <span style='color:#111;'> 3.24KB </span>","children":null,"spread":false},{"title":"README.md <span style='color:#111;'> 6.34KB </span>","children":null,"spread":false},{"title":"fig","children":[{"title":"BrokenNet_filter.png <span style='color:#111;'> 153.71KB </span>","children":null,"spread":false},{"title":"prune_every_layer_by_l1_norm.png <span style='color:#111;'> 565.15KB </span>","children":null,"spread":false},{"title":"quantization_hori.png <span style='color:#111;'> 134.96KB </span>","children":null,"spread":false},{"title":"im2row-crop.png <span style='color:#111;'> 1.44MB </span>","children":null,"spread":false},{"title":"groupSparse.png <span style='color:#111;'> 223.84KB </span>","children":null,"spread":false},{"title":"network_simple.png <span style='color:#111;'> 136.20KB </span>","children":null,"spread":false},{"title":"demo3.png <span style='color:#111;'> 219.42KB </span>","children":null,"spread":false},{"title":"Unrolling the convolution.png <span style='color:#111;'> 351.01KB </span>","children":null,"spread":false},{"title":"shortcut.png <span style='color:#111;'> 139.45KB </span>","children":null,"spread":false},{"title":"greedy.png <span style='color:#111;'> 33.72KB </span>","children":null,"spread":false},{"title":"ill.png <span style='color:#111;'> 98.55KB </span>","children":null,"spread":false},{"title":"theory.png <span style='color:#111;'> 129.09KB </span>","children":null,"spread":false},{"title":"figure_pipeline.png <span style='color:#111;'> 125.12KB </span>","children":null,"spread":false},{"title":"jump.png <span style='color:#111;'> 77.48KB </span>","children":null,"spread":false},{"title":"group_gbn.png <span style='color:#111;'> 111.07KB </span>","children":null,"spread":false},{"title":"Runtime Neural Pruning.png <span style='color:#111;'> 203.82KB </span>","children":null,"spread":false},{"title":"pruning1.png <span style='color:#111;'> 66.20KB </span>","children":null,"spread":false},{"title":"1a2.png <span style='color:#111;'> 98.69KB </span>","children":null,"spread":false},{"title":"Inbound.png <span style='color:#111;'> 139.18KB </span>","children":null,"spread":false}],"spread":false},{"title":"stripe-wise-pruning.pptx <span style='color:#111;'> 84.25KB </span>","children":null,"spread":false}],"spread":true}],"spread":true}]

评论信息

免责申明

【只为小站】的资源来自网友分享,仅供学习研究,请务必在下载后24小时内给予删除,不得用于其他任何用途,否则后果自负。基于互联网的特殊性,【只为小站】 无法对用户传输的作品、信息、内容的权属或合法性、合规性、真实性、科学性、完整权、有效性等进行实质审查;无论 【只为小站】 经营者是否已进行审查,用户均应自行承担因其传输的作品、信息、内容而可能或已经产生的侵权或权属纠纷等法律责任。
本站所有资源不代表本站的观点或立场,基于网友分享,根据中国法律《信息网络传播权保护条例》第二十二条之规定,若资源存在侵权或相关问题请联系本站客服人员,zhiweidada#qq.com,请把#换成@,本站将给予最大的支持与配合,做到及时反馈和处理。关于更多版权及免责申明参见 版权及免责申明