京哥搞异构AI-CSDN博客

原创 [京哥读论文]之ABC-Net：Towards Accurate Binary Convolutional Neural Network

论文题目：Towards Accurate Binary Convolutional Neural Network_NIPS2017引用量:98code:https://github.com/layog/Accurate-Binary-Convolution-Networkhttps://github.com/cow8/ABC-Net-pytorch由于之前二值化工作在ImageNet数据...

2019-08-15 23:08:48 737

原创 [京哥读论文]之Bi-Real Net：Enhancing the Performance of 1-bit CNNs With Improved Representational Capabilit

论文题目：Bi-Real Net: Enhancing the Performance of 1-bit CNNs With Improved Representational Capability and Advanced Training Algorithm_ECCV2018引用量：27code:caffe版：https://github.com/liuzechun/Bi-Real-net...

2019-08-15 23:07:33 705

原创 [京哥读论文]之XNOR再读感想

论文题目：XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks_ECCV2016引用量:1346code:https://github.com/allenai/XNOR-Net,https://github.com/jiecaoyu/XNOR-Net-PyTorch这篇论文主要提出了两个网络...

2019-08-15 23:05:59 459

原创 [京哥读论文]之BNN：Training deep neural networks with weights and activations constrained to+ 1 or-1

论文题目: Binarized neural networks: Training deep neural networks with weights and activations constrained to+ 1 or-1引用量：908code: https://github.com/itayhubara/BinaryNetBNN作者与BinaryConnect是同一个人，还有一个论文...

2019-08-14 14:46:16 1292

原创 [京哥读论文]之BinaryConnect：Training Deep Neural Networks with binary weights during propagations

论文题目:BinaryConnect: Training Deep Neural Networks with binary weights during propagations_NIPS2015引用量：927code:https://github.com/MatthieuCourbariaux/BinaryConnect这是我看的第一篇提出在DNN的前向和反向训练中用1bit的二值权重替代...

2019-08-14 14:43:56 739 1

原创 [京哥读论文]之XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks

Abstract本文是一篇经典的二值化weight和activations的文章，发表在ECCV2016.本文提出了两种有效二值化的框架：XNOR−NetXNOR−NetXNOR-Net以及BWNBWNBWN(Binary-Weight-Networks).在存储方面可以节省32倍的memory。在XNOR−NetXNOR−NetXNOR-Net上weights以及卷积层的input都是二值化...

2018-06-26 22:09:08 648

原创神经网络以及机器学习基础一些链接(不断更新)

BP算法浅显易懂:http://jermmy.xyz/2017/06/25/2017-6-25-reading-notes-neuralnetworksanddeeplearning-2/

2018-06-25 11:08:17 275

原创 [京哥读论文]之From Hashing to CNNs: Training Binary Weight Networks via Hashing

Abstract本文在二值化权重(BWN)方面做出了创新，发表在AAAI2018上，作者是自动化所程建团队。本文的主要贡献是提出了一个新的训练BWN的方法，揭示了哈希与BW(Binary Weights)之间的关联，表明训练BWN的方法在本质上可以当做一个哈希问题。基于这个方法，本文还提出了一种交替更新的方法来有效的学习hash codes而不是直接学习Weights。在小数据和大数据集上表现...

2018-06-05 09:10:35 1143

原创 [京哥读论文]之Extremely Low Bit Neural Network: Squeeze the Last Bit Out with ADMM

Extremely Low Bit Neural Network: Squeeze the Last Bit Out with ADMMAbstract在DeepLearning日益发展的同时，模型压缩的关注度也越来越大。继BWN和TWN之后，这篇论文在超低比特量化领域的一篇新思想文章，发表在AAAI2018上，作者是阿里的。该文的主要思想是将超低bit quantizatio...

2018-06-01 15:51:17 1492

京哥搞异构AI