УДК: 004.931'1, 004.93'14

Extraction of object hierarchy data from trained deep-learning neural networks via analysis of the confusion matrix

Publication in Journal of Optical Technology

For Russian citation (Opticheskii Zhurnal):

Малашин Р.О. Извлечение информации об иерархии объектов из обученных нейронных сетей глубокого обучения с помощью анализа матрицы неточностей // Оптический журнал. 2016. Т. 83. № 10. С. 24–30.

Malashin R.O. Extraction of object hierarchy data from trained deep-learning neural networks via analysis of the confusion matrix [in Russian] // Opticheskii Zhurnal. 2016. V. 83. № 10. P. 24–30.

For citation (Journal of Optical Technology):

R. O. Malashin, "Extraction of object hierarchy data from trained deep-learning neural networks via analysis of the confusion matrix," Journal of Optical Technology. 83(10), 599-603 (2016). https://doi.org/10.1364/JOT.83.000599

Abstract:

We studied the possibility of extracting object hierarchy information from a trained neural network by analyzing the errors obtained on a test sample using an approach based on singular value decomposition of the confusion matrix. Experiments indicate that the methods investigated in this paper can be used to obtain a tentative clustering of classes. In addition, we show that the number of connections within a fully connected layer of a convolutional neural network can be reduced without adversely affecting recognition accuracy for a test sample using locally connected layers. At the same time, however, our experiments did not show that a layer organization consistent with the object hierarchy led to any improvement of the results.

Keywords:

convolutional neural networks, clustering, image recognition

Acknowledgements:

The research was supported by the Ministry of Education and Science of the Russian Federation (Minobrnauka); Russian Federation Government (Subsidy 074-U01).

OCIS codes: 100.4996

References:

1. R. Malashin and A. Kadykov, “Investigation of the generalizing capabilities of convolutional neural networks in forming rotation-invariant attributes,” J. Opt. Technol. 82(8), 509–515 (2015) [Opt. Zh. 82(8), 24–32 (2015)].
2. A. Potapov, V. Batishcheva, and M. Peterson, “Limited generalization capabilities of autoencoders with logistic regression on training sets of small sizes,” in IFIP International Conference on Artificial Intelligence Applications and Innovations (Springer, New York, 2014), pp. 256–264.
3. O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg, and F. F. Li, “ImageNet large scale visual recognition challenge,” Int. J. Comput. Vis. 115(3), 211–252 (2015).
4. A. Krizhevsky, “Learning multiple layers of features from tiny images,” 2009, http://www.cs.toronto.edu/~kriz/learning‑features‑2009‑TR.pdf (accessed 15.03.2016).
5. R. Malashin, V. Lutsiv, A. Kadykov, and N. Degotinskiı˘, “Fast content-based indexing of images,” in Almanac of Scientific Papers from Young Scientists at the 43rd Scientific and Instructional Methodology Conference at ITMO University (2014), pp. 264–266.
6. M. Everingham, L. Van Gool, C. K. Williams, J. Winn, and A. Zisserman, “The Pascal visual object classes (VOC) challenge,” Int. J. Comput. Vis. 88(2), 303–338 (2010).
7. P. Felzenszwalb, R. Girshick, D. McAllester, and D. Ramanan, “Visual object detection with deformable part models,” Commun. ACM 56(9), 97–105 (2013).
8. P. F. Felzenszwalb, R. B. Girshick, D. McAllester, and D. Ramanan, “Object detection with discriminatively trained part-based models,” IEEE Trans. Pattern Anal. Mach. Intell. 32(9), 1627–1645 (2010).
9. A. M. Saxe, J. L. McClelland, and S. Ganguli, “Learning hierarchical category structure in deep neural networks,” in Proceedings of the 35th Annual Meeting of the Cognitive Science Society (2013), pp. 1271–1276.
10. A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet classification with deep convolutional neural networks,” in Advances in Neural Information Processing Systems (2012), pp. 1097–1105.
11. J. Ba and R. Caruana, “Do deep nets really need to be deep?” in Advances in Neural Information Processing Systems (2014), pp. 2654–2662.
12. Caffe, deep learning framework by the BVLC, http://caffe.berkeleyvision.org/ (accessed 14.04. 2016).