Unsupervised Deep Learning by Neighbourhood Discovery

Jiabo Huang, Qi Dong, Shaogang Gong, Xiatian Zhu
Proceedings of the 36th International Conference on Machine Learning, PMLR 97:2849-2858, 2019.

Abstract

Deep convolutional neural networks (CNNs) have demonstrated remarkable success in computer vision by supervisedly learning strong visual feature representations. However, training CNNs relies heavily on the availability of exhaustive training data annotations, limiting significantly their deployment and scalability in many application scenarios. In this work, we introduce a generic unsupervised deep learning approach to training deep models without the need for any manual label supervision. Specifically, we progressively discover sample anchored/centred neighbourhoods to reason and learn the underlying class decision boundaries iteratively and accumulatively. Every single neighbourhood is specially formulated so that all the member samples can share the same unseen class labels at high probability for facilitating the extraction of class discriminative feature representations during training. Experiments on image classification show the performance advantages of the proposed method over the state-of-the-art unsupervised learning models on six benchmarks including both coarse-grained and fine-grained object image categorisation.

Cite this Paper

BibTeX
@InProceedings{pmlr-v97-huang19b, title = {Unsupervised Deep Learning by Neighbourhood Discovery}, author = {Huang, Jiabo and Dong, Qi and Gong, Shaogang and Zhu, Xiatian}, booktitle = {Proceedings of the 36th International Conference on Machine Learning}, pages = {2849--2858}, year = {2019}, editor = {Chaudhuri, Kamalika and Salakhutdinov, Ruslan}, volume = {97}, series = {Proceedings of Machine Learning Research}, month = {09--15 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v97/huang19b/huang19b.pdf}, url = {https://proceedings.mlr.press/v97/huang19b.html}, abstract = {Deep convolutional neural networks (CNNs) have demonstrated remarkable success in computer vision by supervisedly learning strong visual feature representations. However, training CNNs relies heavily on the availability of exhaustive training data annotations, limiting significantly their deployment and scalability in many application scenarios. In this work, we introduce a generic unsupervised deep learning approach to training deep models without the need for any manual label supervision. Specifically, we progressively discover sample anchored/centred neighbourhoods to reason and learn the underlying class decision boundaries iteratively and accumulatively. Every single neighbourhood is specially formulated so that all the member samples can share the same unseen class labels at high probability for facilitating the extraction of class discriminative feature representations during training. Experiments on image classification show the performance advantages of the proposed method over the state-of-the-art unsupervised learning models on six benchmarks including both coarse-grained and fine-grained object image categorisation.} }
Endnote
%0 Conference Paper %T Unsupervised Deep Learning by Neighbourhood Discovery %A Jiabo Huang %A Qi Dong %A Shaogang Gong %A Xiatian Zhu %B Proceedings of the 36th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2019 %E Kamalika Chaudhuri %E Ruslan Salakhutdinov %F pmlr-v97-huang19b %I PMLR %P 2849--2858 %U https://proceedings.mlr.press/v97/huang19b.html %V 97 %X Deep convolutional neural networks (CNNs) have demonstrated remarkable success in computer vision by supervisedly learning strong visual feature representations. However, training CNNs relies heavily on the availability of exhaustive training data annotations, limiting significantly their deployment and scalability in many application scenarios. In this work, we introduce a generic unsupervised deep learning approach to training deep models without the need for any manual label supervision. Specifically, we progressively discover sample anchored/centred neighbourhoods to reason and learn the underlying class decision boundaries iteratively and accumulatively. Every single neighbourhood is specially formulated so that all the member samples can share the same unseen class labels at high probability for facilitating the extraction of class discriminative feature representations during training. Experiments on image classification show the performance advantages of the proposed method over the state-of-the-art unsupervised learning models on six benchmarks including both coarse-grained and fine-grained object image categorisation.
APA
Huang, J., Dong, Q., Gong, S. & Zhu, X.. (2019). Unsupervised Deep Learning by Neighbourhood Discovery. Proceedings of the 36th International Conference on Machine Learning, in Proceedings of Machine Learning Research 97:2849-2858 Available from https://proceedings.mlr.press/v97/huang19b.html.

Related Material