Unsupervised Learning on Neural Network Outputs: with Application in Zero-shot Learning

Yao Lu

Unsupervised Learning on Neural Network Outputs: with Application in Zero-shot Learning

Yao Lu

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

Abstract

The outputs of a trained neural network contain much richer information than just a one-hot classifier. For example, a neural network might give an image of a dog the probability of one in a million of being a cat but it is still much larger than the probability of being a car. To reveal the hidden structure in them, we apply two unsupervised learning algorithms, PCA and ICA, to the outputs of a deep Convolutional Neural Network trained on the ImageNet of 1000 classes. The PCA/ICA embedding of the object classes reveals their visual similarity and the PCA/ICA components can be interpreted as common visual features shared by similar object classes. For an application, we proposed a new zero-shot learning method, in which the visual features learned by PCA/ICA are employed. Our zeroshot learning method achieves the state-of-the-art results on the ImageNet of over 20000 classes.

Original language	English
Title of host publication	Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence
Subtitle of host publication	New York, New York, USA, 9–15 July 2016
Editors	Subbarao Kambhampati
Publisher	AAAI PRESS
Pages	3432-3428
ISBN (Electronic)	978-1-57735-770-4
Publication status	Published - 9 Jul 2016
MoE publication type	A4 Article in a conference publication
Event	International Joint Conference on Artificial Intelligence - New York Hilton Midtown, New York, United States Duration: 9 Jul 2016 → 15 Jul 2016 Conference number: 25 http://ijcai-16.org/index.php/welcome/view/home

Publication series

Name	International Joint Conferences on Artificial Intelligence
Publisher	IAAA press
ISSN (Electronic)	1045-0823

Conference

Conference	International Joint Conference on Artificial Intelligence
Abbreviated title	IJCAI
Country	United States
City	New York
Period	09/07/2016 → 15/07/2016
Internet address	http://ijcai-16.org/index.php/welcome/view/home

Fingerprint Dive into the research topics of 'Unsupervised Learning on Neural Network Outputs: with Application in Zero-shot Learning'. Together they form a unique fingerprint.

Cite this

@inproceedings{29e9168e356b408b93e9438705a8cd52,

title = "Unsupervised Learning on Neural Network Outputs: with Application in Zero-shot Learning",

abstract = "The outputs of a trained neural network contain much richer information than just a one-hot classifier. For example, a neural network might give an image of a dog the probability of one in a million of being a cat but it is still much larger than the probability of being a car. To reveal the hidden structure in them, we apply two unsupervised learning algorithms, PCA and ICA, to the outputs of a deep Convolutional Neural Network trained on the ImageNet of 1000 classes. The PCA/ICA embedding of the object classes reveals their visual similarity and the PCA/ICA components can be interpreted as common visual features shared by similar object classes. For an application, we proposed a new zero-shot learning method, in which the visual features learned by PCA/ICA are employed. Our zeroshot learning method achieves the state-of-the-art results on the ImageNet of over 20000 classes.",

author = "Yao Lu",

year = "2016",

month = jul,

day = "9",

language = "English",

series = "International Joint Conferences on Artificial Intelligence",

publisher = "AAAI PRESS",

pages = "3432--3428",

editor = "Subbarao Kambhampati",

booktitle = "Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence",

note = "International Joint Conference on Artificial Intelligence, IJCAI ; Conference date: 09-07-2016 Through 15-07-2016",

url = "http://ijcai-16.org/index.php/welcome/view/home",

}

Lu, Y 2016, Unsupervised Learning on Neural Network Outputs: with Application in Zero-shot Learning. in S Kambhampati (ed.), Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence: New York, New York, USA, 9–15 July 2016. International Joint Conferences on Artificial Intelligence, AAAI PRESS, pp. 3432-3428, International Joint Conference on Artificial Intelligence, New York, United States, 09/07/2016.

Unsupervised Learning on Neural Network Outputs: with Application in Zero-shot Learning. / Lu, Yao.

Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence: New York, New York, USA, 9–15 July 2016. ed. / Subbarao Kambhampati. AAAI PRESS, 2016. p. 3432-3428 (International Joint Conferences on Artificial Intelligence).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

TY - GEN

T1 - Unsupervised Learning on Neural Network Outputs: with Application in Zero-shot Learning

AU - Lu, Yao

N1 - Conference code: 25

PY - 2016/7/9

Y1 - 2016/7/9

N2 - The outputs of a trained neural network contain much richer information than just a one-hot classifier. For example, a neural network might give an image of a dog the probability of one in a million of being a cat but it is still much larger than the probability of being a car. To reveal the hidden structure in them, we apply two unsupervised learning algorithms, PCA and ICA, to the outputs of a deep Convolutional Neural Network trained on the ImageNet of 1000 classes. The PCA/ICA embedding of the object classes reveals their visual similarity and the PCA/ICA components can be interpreted as common visual features shared by similar object classes. For an application, we proposed a new zero-shot learning method, in which the visual features learned by PCA/ICA are employed. Our zeroshot learning method achieves the state-of-the-art results on the ImageNet of over 20000 classes.

AB - The outputs of a trained neural network contain much richer information than just a one-hot classifier. For example, a neural network might give an image of a dog the probability of one in a million of being a cat but it is still much larger than the probability of being a car. To reveal the hidden structure in them, we apply two unsupervised learning algorithms, PCA and ICA, to the outputs of a deep Convolutional Neural Network trained on the ImageNet of 1000 classes. The PCA/ICA embedding of the object classes reveals their visual similarity and the PCA/ICA components can be interpreted as common visual features shared by similar object classes. For an application, we proposed a new zero-shot learning method, in which the visual features learned by PCA/ICA are employed. Our zeroshot learning method achieves the state-of-the-art results on the ImageNet of over 20000 classes.

M3 - Conference contribution

T3 - International Joint Conferences on Artificial Intelligence

SP - 3432

EP - 3428

BT - Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence

A2 - Kambhampati, Subbarao

PB - AAAI PRESS

T2 - International Joint Conference on Artificial Intelligence

Y2 - 9 July 2016 through 15 July 2016

ER -