Filters








3,934 Hits in 5.5 sec

Empirical Evaluation of Whitening and Optimization of Feature Learning

Nouman Qadeer, Xiabi Liu
<span title="">2014</span> <i title="IEEE"> <a target="_blank" rel="noopener" href="https://fatcat.wiki/container/r2ok75u3w5a3vhsdwqc4z573he" style="color: black;">2014 UKSim-AMSS 16th International Conference on Computer Modelling and Simulation</a> </i> &nbsp;
Deep learning and Feature learning emerged new field in machine learning and beat many, state of the arts results in diverse areas.  ...  Different Whitening preprocessing techniques and optimization methods were applied on well known data set corel-100 and found out that Cost effective PCA whitening is also same reliable as cost prone other  ...  OPTIMIZATION Optimization remains a well studied area of machine learning.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.1109/uksim.2014.77">doi:10.1109/uksim.2014.77</a> <a target="_blank" rel="external noopener" href="https://dblp.org/rec/conf/uksim/QadeerL14.html">dblp:conf/uksim/QadeerL14</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/vm2ct53pzbg53boxyraraevtuq">fatcat:vm2ct53pzbg53boxyraraevtuq</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20170810081720/http://ijssst.info/Vol-15/No-4/data/4923a036.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/c4/1b/c41b7488b082e9b72c9c32c7fe9ef91e26c40137.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.1109/uksim.2014.77"> <button class="ui left aligned compact blue labeled icon button serp-button"> <i class="external alternate icon"></i> ieee.com </button> </a>

Improving Generalization of Batch Whitening by Convolutional Unit Optimization [article]

Yooshin Cho, Hanbyel Cho, Youngsoo Kim, Junmo Kim
<span title="2021-11-02">2021</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
Notably, we verify that our method improves stability and performance of whitening when using large learning rate, group size, and iteration number.  ...  In commonly used structures, which are empirically optimized with Batch Normalization, the normalization layer appears between convolution and activation function.  ...  Our Convolutional Unit significantly stabilizes whitening modules by increasing the rank of features, and improves efficacy by properly choosing the target of whitening and removing the linear transform  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2108.10629v2">arXiv:2108.10629v2</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/nwsjt2d7ezakrkqfee7scwufau">fatcat:nwsjt2d7ezakrkqfee7scwufau</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20211108140029/https://arxiv.org/pdf/2108.10629v2.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/6b/a4/6ba44a4ef34db0819a95046d76e398723a4ab3a3.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2108.10629v2" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

An Investigation into the Stochasticity of Batch Whitening [article]

Lei Huang, Lei Zhao, Yi Zhou, Fan Zhu, Li Liu, Ling Shao
<span title="2020-03-27">2020</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
We quantitatively investigate the stochasticity of different whitening transformations and show that it correlates well with the optimization behaviors during training.  ...  A full understanding of the process has been a central target in the deep learning communities.  ...  Acknowledgement We thank Anna Hennig and Ying Hu for their help with proofreading.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2003.12327v1">arXiv:2003.12327v1</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/7m3aoov6rvhgpfpvzxduk5dssi">fatcat:7m3aoov6rvhgpfpvzxduk5dssi</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200903194716/https://arxiv.org/pdf/2003.12327v1.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/76/b2/76b2f165805e3c8e57be7c80f0adf078f5f83573.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2003.12327v1" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Group Whitening: Balancing Learning Efficiency and Representational Capacity [article]

Lei Huang, Yi Zhou, Li Liu, Fan Zhu, Ling Shao
<span title="2021-04-06">2021</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
This paper proposes group whitening (GW), which exploits the advantages of the whitening operation and avoids the disadvantages of normalization within mini-batches.  ...  The merits of BN in improving a model's learning efficiency can be further amplified by applying whitening, while its drawbacks in estimating population statistics for inference can be avoided through  ...  We train over 160 epochs and divide the learning rate by 5 at 60 and 120 epochs. We evaluate the best training accuracy among the learning rates of {0.001, 0.01, 0.05, 0.1, 0.5}.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2009.13333v4">arXiv:2009.13333v4</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/7ahb5vtfibcpnattgwrgpw3eem">fatcat:7ahb5vtfibcpnattgwrgpw3eem</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20210408072504/https://arxiv.org/pdf/2009.13333v4.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/08/29/0829440d1616343f7e48896cfcf3e41241e7e0ae.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2009.13333v4" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Network topology of turbulent premixed Bunsen flame at elevated pressure and turbulence intensity

Jinhua Wang, Yaohui Nie, Weijie Zhang, Shilong Guo, Meng Zhang, Zuohua Huang
<span title="">2019</span> <i title="Elsevier BV"> <a target="_blank" rel="noopener" href="https://fatcat.wiki/container/iyityt2owngmhg4uesksbov7ge" style="color: black;">Aerospace Science and Technology</a> </i> &nbsp;
Empirical evaluation implies that our approach learns more informative feature mappings and is more efficient to optimize.  ...  The objective of Soft-HGR is straightforward, only involving two inner products, which guarantees the efficiency and stability in optimization.  ...  Acknowledgement The research of Shao-Lun Huang was funded by the Natural Science Foundation of China 61807021, and Shenzhen Municipal Scientific Program JCYJ20170818094022586.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.1016/j.ast.2019.105361">doi:10.1016/j.ast.2019.105361</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/vifdxmg6bfb6dhconc7i7c5khy">fatcat:vifdxmg6bfb6dhconc7i7c5khy</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200305122237/https://www.aaai.org/ojs/index.php/AAAI/article/download/4464/4342" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/4b/aa/4baa0ec214d2676a8d04ec37e594bce388e72f8f.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.1016/j.ast.2019.105361"> <button class="ui left aligned compact blue labeled icon button serp-button"> <i class="external alternate icon"></i> elsevier.com </button> </a>

An Efficient Approach to Informative Feature Extraction from Multimodal Data [article]

Lichen Wang, Jiaxiang Wu, Shao-Lun Huang, Lizhong Zheng, Xiangxiang Xu, Lin Zhang, Junzhou Huang
<span title="2019-09-21">2019</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
Empirical evaluation implies that our approach learns more informative feature mappings and is more efficient to optimize.  ...  The objective of Soft-HGR is straightforward, only involving two inner products, which guarantees the efficiency and stability in optimization.  ...  Acknowledgement The research of Shao-Lun Huang was funded by the Natural Science Foundation of China 61807021, and Shenzhen Municipal Scientific Program JCYJ20170818094022586.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1811.08979v2">arXiv:1811.08979v2</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/wtpyzd3fmvgrjaorfciftgoxmi">fatcat:wtpyzd3fmvgrjaorfciftgoxmi</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200915134629/https://arxiv.org/pdf/1811.08979v2.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/06/75/06757225454e939a880abdc1bd64f818460dd107.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1811.08979v2" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Are sparse representations really relevant for image classification?

Roberto Rigamonti, Matthew A. Brown, Vincent Lepetit
<span title="">2011</span> <i title="IEEE"> <a target="_blank" rel="noopener" href="https://fatcat.wiki/container/ilwxppn4d5hizekyd3ndvy2mii" style="color: black;">CVPR 2011</a> </i> &nbsp;
In this paper we evaluate its impact on the recognition rate using a shallow modular architecture, adopting both standard filter banks and filter banks learned in an unsupervised way.  ...  Recent years have seen an increasing interest in sparse representations for image classification and object recognition, probably motivated by evidence from the analysis of the primate visual cortex.  ...  In this paper we aim to evaluate the actual importance of sparsity in image classification, by performing an extensive empirical evaluation and adopting the recognition rate as a criterion.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.1109/cvpr.2011.5995313">doi:10.1109/cvpr.2011.5995313</a> <a target="_blank" rel="external noopener" href="https://dblp.org/rec/conf/cvpr/RigamontiBL11.html">dblp:conf/cvpr/RigamontiBL11</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/ibiuooaahjep5pz3ulr3ooeea4">fatcat:ibiuooaahjep5pz3ulr3ooeea4</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20170818071856/https://infoscience.epfl.ch/record/167170/files/rigamonti_cvpr2011.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/f0/16/f0166ff01dc59df880a182bf46ec35d257d9e8ba.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.1109/cvpr.2011.5995313"> <button class="ui left aligned compact blue labeled icon button serp-button"> <i class="external alternate icon"></i> ieee.com </button> </a>

Whitening and second order optimization both make information in the dataset unusable during training, and can reduce or prevent generalization [article]

Neha S. Wadia, Daniel Duckworth, Samuel S. Schoenholz, Ethan Dyer, Jascha Sohl-Dickstein
<span title="2021-07-19">2021</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
We show that both data whitening and second order optimization can harm or entirely prevent generalization.  ...  Machine learning is predicated on the concept of generalization: a model achieving low error on a sufficiently large training set should also perform well on novel samples from the same distribution.  ...  Acknowledgements We thank Jeffrey Pennington for help formulating the project, and Justin Gilmer, Roger Grosse, Nicolas Le Roux, and Jesse Livezey for detailed feedback on a manuscript draft.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2008.07545v4">arXiv:2008.07545v4</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/xqwvfhzvlzbxbdk4vhpmd3caue">fatcat:xqwvfhzvlzbxbdk4vhpmd3caue</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200901050408/https://arxiv.org/pdf/2008.07545v2.pdf" title="fulltext PDF download [not primary version]" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <span style="color: #f43e3e;">&#10033;</span> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2008.07545v4" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Universal Style Transfer via Feature Transforms [article]

Yijun Li, Chen Fang, Jimei Yang, Zhaowen Wang, Xin Lu, Ming-Hsuan Yang
<span title="2017-11-17">2017</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
The whitening and coloring transforms reflect a direct matching of feature covariance of the content image to a given style image, which shares similar spirits with the optimization of Gram matrix based  ...  The key ingredient of our method is a pair of feature transforms, whitening and coloring, that are embedded to an image reconstruction network.  ...  Acknowledgments This work is supported in part by the NSF CAREER Grant #1149783, gifts from Adobe and NVIDIA.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1705.08086v2">arXiv:1705.08086v2</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/eemmpv5cgjcqllsfj6i6tx2sx4">fatcat:eemmpv5cgjcqllsfj6i6tx2sx4</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200827201510/https://arxiv.org/pdf/1705.08086v2.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/74/d2/74d251760b912ce33b3680f7e57b70a41cebe7a0.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1705.08086v2" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

On Feature Decorrelation in Self-Supervised Learning [article]

Tianyu Hua, Wenxiao Wang, Zihui Xue, Sucheng Ren, Yue Wang, Hang Zhao
<span title="2021-08-25">2021</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
The gains from feature decorrelation are verified empirically to highlight the importance and the potential of this insight.  ...  In self-supervised representation learning, a common idea behind most of the state-of-the-art approaches is to enforce the robustness of the representations to predefined augmentations.  ...  Feature Decorrelation on ImageNet The accuracy in linear evaluation on ImageNet has become a de facto metric of visual features learned in selfsupervised fashions.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2105.00470v2">arXiv:2105.00470v2</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/oph5osy5e5cndchu7vgmkzkr7m">fatcat:oph5osy5e5cndchu7vgmkzkr7m</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20210903164447/https://arxiv.org/pdf/2105.00470v2.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/c4/6a/c46af3a9bd7954df136c00698a62c5db8c1634b3.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2105.00470v2" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Feature Transformation with Class Conditional Decorrelation

Xu-Yao Zhang, Kaizhu Huang, Cheng-Lin Liu
<span title="">2013</span> <i title="IEEE"> <a target="_blank" rel="noopener" href="https://fatcat.wiki/container/gckg3mzs4fhxhbrvmbsa54bccm" style="color: black;">2013 IEEE 13th International Conference on Data Mining</a> </i> &nbsp;
We also discuss the potential applications of CCD for other problems such as Gaussian mixture models and classifier ensemble learning.  ...  The well-known feature transformation model of Fisher linear discriminant analysis (FDA) can be decomposed into an equivalent two-step approach: whitening followed by principal component analysis (PCA)  ...  The feature dimensionality is 512. The number of training and testing samples are 898, 573 and 224, 559 respectively, for which the statistical significance of evaluations should be sufficient.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.1109/icdm.2013.43">doi:10.1109/icdm.2013.43</a> <a target="_blank" rel="external noopener" href="https://dblp.org/rec/conf/icdm/ZhangHL13.html">dblp:conf/icdm/ZhangHL13</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/4nf6gafzyzervm6li3yaz27pvi">fatcat:4nf6gafzyzervm6li3yaz27pvi</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20170809015442/http://www.nlpr.ia.ac.cn/pal/xyz/Publication/XYZ2013-class_conditional_decorrelation-ICDM.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/a0/52/a052d2d92771c5c54341a6a3e919c4f33d44a0fb.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.1109/icdm.2013.43"> <button class="ui left aligned compact blue labeled icon button serp-button"> <i class="external alternate icon"></i> ieee.com </button> </a>

Semantic-Aware Domain Generalized Segmentation [article]

Duo Peng, Yinjie Lei, Munawar Hayat, Yulan Guo, Wen Li
<span title="2022-04-02">2022</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
With the help of SAN and SAW, we encourage both intra-category compactness and inter-category separability. We validate our approach through extensive experiments on widely-used datasets (i.e.  ...  GTAV, SYNTHIA, Cityscapes, Mapillary and BDDS). Our approach shows significant improvements over existing state-of-the-art on various backbone networks.  ...  Further Implementation Details We follow previous work [3, 11] to adopt normalization and whitening at the first two stages of convolution layers, since shallow layers encode more style information [11  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2204.00822v1">arXiv:2204.00822v1</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/r6exzrn3dzf5tjpdfw2bddxpbi">fatcat:r6exzrn3dzf5tjpdfw2bddxpbi</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20220406203922/https://arxiv.org/pdf/2204.00822v1.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/e8/b2/e8b2d52d703dd66a460a614348679307692b6147.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2204.00822v1" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Concept Whitening for Interpretable Image Recognition [article]

Zhi Chen, Yijie Bei, Cynthia Rudin
<span title="2020-10-19">2020</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
CW is an alternative to a batch normalization layer in that it normalizes, and also decorrelates (whitens) the latent space.  ...  When a concept whitening module is added to a CNN, the axes of the latent space are aligned with known concepts of interest.  ...  However, a scalar is needed to quantify how much a sample is activated on a concept, which is used in both optimization and evaluation.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2002.01650v4">arXiv:2002.01650v4</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/g3mopaj5gzfvtm5dpprtdejp5u">fatcat:g3mopaj5gzfvtm5dpprtdejp5u</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20201119004340/https://arxiv.org/pdf/2002.01650v4.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/10/4b/104bbdc862c16b97535083ef06dd585bb65e7ca4.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2002.01650v4" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Whitening and Coloring batch transform for GANs [article]

Aliaksandr Siarohin, Enver Sangineto, Nicu Sebe
<span title="2019-02-25">2019</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
Moreover, we show that full-feature whitening is important in a general GAN scenario in which the training process is known to be highly unstable.  ...  In this paper we propose to generalize both BN and cBN using a Whitening and Coloring based batch normalization.  ...  ACKNOWLEDGMENTS We thank the NVIDIA Corporation for the donation of the GPUs used in this project.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1806.00420v2">arXiv:1806.00420v2</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/pzozpdoyk5favmtpk5fdaly7fi">fatcat:pzozpdoyk5favmtpk5fdaly7fi</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200905165806/https://arxiv.org/pdf/1806.00420v2.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/e5/cf/e5cf28063f76ad1a3cda56f68b08ff264a0aa198.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1806.00420v2" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Improving STDP-based Visual Feature Learning with Whitening [article]

Pierre Falez and Pierre Tirilly and Ioan Marius Bilasco
<span title="2020-02-24">2020</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
In this paper, we propose to use whitening as a pre-processing step before learning features with STDP.  ...  We also propose an approximation of whitening as convolution kernels that is computationally cheaper to learn and more suited to be implemented on neuromorphic hardware.  ...  As we aim at evaluating only the ability of STDP to learn visual features, we rely on more classical tools for the feature aggregator f a (max pooling) and the classifier f c (SVM).  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2002.10177v1">arXiv:2002.10177v1</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/jujwy7hfvbdwnam7q5seszqafe">fatcat:jujwy7hfvbdwnam7q5seszqafe</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200321153625/https://arxiv.org/pdf/2002.10177v1.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2002.10177v1" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>
&laquo; Previous Showing results 1 &mdash; 15 out of 3,934 results