Understanding Intra-Class Knowledge Inside CNN

Donglai Wei, Bolei Zhou, Antonio Torrabla, William T. Freeman · arXiv.org · 2015

Summary

To invert the intra-class knowledge inside CNN into more interpretable images, a non-parametric patch prior upon previous CNN visualization models is proposed and it is shown how different "styles" of templates for an object class are organized by CNN in terms of location and content, and represented in a hierarchical and ensemble way.

Type: preprint
Published: 2015-07-09
Cited by: 94
References: 14
Access: Open access
DOI: https://doi.org/10.48550/arxiv.1507.02379
arXiv: https://arxiv.org/abs/1507.02379
OpenAlex: https://openalex.org/W830575572
Semantic Scholar: https://api.semanticscholar.org/CorpusID:10774286

Keywords

Class (philosophy), Computer science, Artificial intelligence

References

Intriguing properties of neural networksChristian Szegedy, Wojciech Zaremba, Ilya Sutskever — 2013
Understanding deep image representations by inverting themAravindh Mahendran, Andrea Vedaldi — 2014
Discovering states and transformations in image collectionsPhillip Isola, Joseph J. Lim, Edward H. Adelson — 2015
PatchMatch: a randomized correspondence algorithm for structural image editingConnelly Barnes, Eli Shechtman, Adam Finkelstein — 2009
ImageNet: A large-scale hierarchical image databaseJia Deng, Wei Dong, Richard Socher — 2009
Gradient-based learning applied to document recognitionYann LeCun, Léon Bottou, Yoshua Bengio — 1998
Do Convnets Learn Correspondence?Jonathan Long, Ning Zhang, Trevor Darrell — 2014
ImageNet classification with deep convolutional neural networksAlex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton — 2012
Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency MapsKaren Simonyan, Andrea Vedaldi, Andrew Zisserman — 2013
Understanding DropoutPierre Baldi, Peter Sadowski — 2013
Analyzing the Performance of Multilayer Neural Networks for Object RecognitionPulkit Agrawal, Ross Girshick, Jitendra Malik — 2014
Object Detectors Emerge in Deep Scene CNNsBolei Zhou, Aditya Khosla, Àgata Lapedriza — 2014
Object Detectors Emerge in Deep Scene CNNsBolei Zhou, Aditya Khosla, Àgata Lapedriza — 2015
Visualizing and Understanding Convolutional NetworksMatthew D. Zeiler, Rob Fergus — 2013
GradientBased Learning Applied to Document RecognitionS. Haykin, B. Kosko — 2001

Cited by

Seeing Behind the Camera: Identifying the Authorship of a PhotographChristopher Thomas, Adriana Kovashka — 2015
Multifaceted Feature Visualization: Uncovering the Different Types of Features Learned By Each Neuron in Deep Neural NetworksAnh Mai Nguyen, Jason Yosinski, Jeff Clune — 2016
Towards Better Analysis of Deep Convolutional Neural NetworksMengchen Liu, Jiaxin Shi, Zhen Li — 2016
A Powerful Generative Model Using Random Weights for the Deep Image RepresentationKun He, Yan Wang, John E. Hopcroft — 2016
Detection and classification of breast cancer in whole slide histopathology images using deep convolutional networksBarış Geçer — 2016
What makes ImageNet good for transfer learning?Minyoung Huh, Pulkit Agrawal, Alexei A. Efros — 2016
Every Filter Extracts A Specific Texture In Convolutional Neural NetworksZhiqiang Xia, Ce Zhu, Zhengtao Wang — 2016
Grandmother cells and localist representations: a review of current thinkingJeffrey S. Bowers — 2017
On the Selective and Invariant Representation of DCNN for High-Resolution Remote Sensing Image RecognitionJie Chen, Chao Yuan, Min Deng — 2017
Do Convolutional Neural Networks Learn Class Hierarchy?Bilal Alsallakh, Amin Jourabloo, Mao Ye — 2017
Visualizing deep neural network by alternately image blurring and deblurringWang Feng, Haijun Liu, Jian Cheng — 2018
One Pixel Attack for Fooling Deep Neural NetworksJiawei Su, Danilo Vasconcellos Vargas, Kouichi Sakurai — 2017
Generative adversarial network for visualizing convolutional networkMasayuki Kobayashi, Masanori Suganuma, Tomoharu Nagao — 2017
On monitoring development using high resolution satellite imagesP. Suraj, Ankesh Gupta, Makkunda Sharma — 2017
Beyond saliency: understanding convolutional neural networks from saliency prediction on layer-wise relevance propagationHeyi Li, Yunke Tian, Klaus Mueller — 2017
Visual Analytics in Deep Learning: An Interrogative Survey for the Next FrontiersFred Hohman, Minsuk Kahng, Robert Pienta — 2018
Understanding CNN via deep features analysisHao Xu, Yueru Chen, Ruiyuan Lin — 2017
Class Subset Selection for Transfer Learning using SubmodularityVarun Manjunatha, Srikumar Ramalingam, Tim K. Marks — 2018
Attacking convolutional neural network using differential evolutionJiawei Su, Danilo Vargas, Kouichi Sakurai — 2018
What Makes a Video a Video: Analyzing Temporal Information in Video Understanding Models and DatasetsDe-An Huang, Vignesh Ramanathan, Dhruv Mahajan — 2018

Understanding Intra-Class Knowledge Inside CNN

Summary

Keywords

References

Cited by

Related papers