Abstract
In this paper, we propose the use of 2D pre-defined convolutional layers based on parabolic and hyperbolic Partial Differential Equations (PDEs) to learn low-level abstractions. We demonstrate their effectiveness by adding them on top of ResNet, ResNext, DenseNet, and VGG architectures and performing numerical experiments on the CIFAR-10 dataset.
Similar content being viewed by others
REFERENCES
Bengio, Y., Learning deep architectures for AI, Found. Trends Mach. Learn., 2009, vol. 2, no. 1, pp. 1–155.
LeCun, Y., Bengio, Y., and Hinton, G., Deep learning, Nature, 2015, vol. 521, p. 436.
Aubert, G. and Kornprobst, P., Mathematical Problems in Image Processing: Partial Differential Equations and the Calculus of Variations, New York: Springer Verlag, 2002.
He, K., Zhang, X., Ren, S., and Sun, J., Deep residual learning for image recognition, 2015. arXiv: 1512.03385v1 [cs.CV].
Xie, S., Girshick, R., Dollár, P., Tu, Z., and He, K., Aggregated residual transformations for deep neural networks, Proc. Conf. on Computer Vision and Pattern Recognition, CVPR, Honolulu, 2017.
Huang, G., Liu, Z., Maaten, L., and Weinberger, K.Q., Densely connected convolutional networks, Proc. Conf. on Computer Vision and Pattern Recognition, CVPR, Honolulu, 2017.
Ruthotto, L. and Haber, E., Deep neural networks motivated by partial differential equations, 2019. arXiv:1804.04272v2 [cs.LG].
Kag, A. and Saligrama, V., Condensing CNNs with partial differential equation, Proc. Computer Vision and Pattern Recognition Conf., New Orleans, 2022.
Guo, P., Huang, K., and Xu, Z., Partial differential equations is all you need for generating neural architectures, 2021. arXiv:2103.08313v1 [cs.LG].
Kundu, S., Nazemi, M., Pendram, M., Chugg, K.M., and Beerel, P.A., Pre-defined sparsity for low-complexity convolutional neural networks, IEEE Trans. Comput., 2020, vol. 99, p. 1–1.
Ma, Y., Luo, Y., and Yang, Z., PCFNet: Deep neural network with predefined convolutional filters, Neurocomputing, 2020, vol. 382, no. 7.
Richtmyer, R.D. and Morton, K.W., Difference Methods for Initial-Value Problems, 2nd ed., New York, London, Sydney: Interscience Publ. John Wiley & Sons, 1967, pp. 1–30.
Dong, S., Finite Difference Methods for the Hyperbolic Wave Partial Differential Equations, Princeton Plasma Physics Laboratory, 2006, pp. 1–17.
Krizhevsky, A., Learning multiple layers of features from tiny images, Sci. Tech. Rep., Univ. of Toronto, Dep. of Computer, 2009.
Lee, C.Y., Xie, S., Gallagher, P., Zhang, Z., and Tu, Z., Deeply-supervised nets, 2014. arXiv:1409.5185.
Simonyan, K. and Zisserman, A., Very deep convolutional networks for large-scale image recognition, Proc. Int. Conf. on Learning Representations, ICLR, San Diego, 2015.
Gabor, D., Very theory of communication. Part 1: the analysis of information, J. Inst. Electr. Eng.-Part III: Radio Commun. Eng., 1946, vol. 93, no. 26, pp. 429–441.
Schmid, C., Constructing models for content-based image retrieval, Proc. IEEE Computer Society Conf. on Computer Vision and Pattern Recognition, CVPR, Kauai, HI, 2001.
Gao, W., Zhang, X., Yang, L., and Liu, H., An improved sobel edge detection, Proc. 3rd IEEE Int. Conf. on Conference: Computer Science and Information Technology (ICCSIT), Chengdu, 2010, vol. 5, pp. 67–71.
ACKNOWLEDGMENTS
The work was supported by the Science Committee of RA, in the frames of the research project no. 23DP-1B017. Also, the authors wish to thank Professor Armenak Babayan and Tsolak Ghukasyan for their assistance and advice.
Author information
Authors and Affiliations
Corresponding authors
Ethics declarations
The authors declare that they have no conflicts of interest.
Rights and permissions
About this article
Cite this article
Sahakyan, V.R., Melkonyan, V.G., Gharagyozyan, G.A. et al. Enhancing Image Recognition with Pre-Defined Convolutional Layers Based on PDEs. Program Comput Soft 49, 192–197 (2023). https://doi.org/10.1134/S0361768823030088
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1134/S0361768823030088