Comparison of compressed vector representations of matrices using compartment spiking neuron model (CSNM)

Comparison of compressed vector representations of matrices using compartment spiking neuron model (CSNM)

Ivan S. Fomin
Russian State Scientific Center for Robotics and Technical Cybernetics (RTC), Junior Research Scientist, 21, Tikhoretsky pr., Saint Petersburg, 194064, Russia, This email address is being protected from spambots. You need JavaScript enabled to view it., ORCID: 0000-0001-9066-4836

Received September 13, 2024

Abstract
In this paper, we consider the problem of comparing compressed vector representations of matrices using a spike neuron. Some parameters of matrices proximity are known, some of them a system must determine during the setup and training process. In the second option, you need to compare the remaining part of the group with the one known at the time of training. Compression is performed using a siamese convolutional neural network prepared in a special way. To compare points in the vector representation space, it is proposed to use the segment spiking neuron model (CSNM), which has proven itself well in other similar tasks. Simple mathematical operations are used to convert a point into a spike representation suitable for classification by a segment spiking neuron. The description of the criteria for choosing the network architecture and the results of the selection are given. A variant of a siamese convolutional network based on the well-known ResNet-18 architecture and a similar set of experiments using it is also presented. A way to accelerate learning by selecting complex training examples is shown. The results demonstrate the applicability of the proposed approaches to solving the problem of comparing vector representations. The proposed method shows a quality of about 92-95% when solving the comparison problem for the first task, and about 70% when solving the comparison problem of the remaining part of the known. In particular, the approach may be of interest for civilian applications – comparing transport passengers, accounting for access and video surveillance in the security systems of thermal and nuclear power plants, recognition of visitors to automated stores, etc., if some information about the object (or person) is presented in matrix form.

Key words
Matrix transformation; vector comparison; compartmental spiking neuron model; siamese neural network; convolutional neural network;  object recognition; object classification.

Acknowledgments
The work was carried out as the part of the state task of the Russian Ministry of Education and Science «Research of methods for creating self-learning video surveillance systems and video analytics based on the integration of technologies for spatiotemporal filtering of video stream and neural networks» (FNRG 2022 0015 1021060307687-9-1.2.1 №075-00697-24-00 from 27.12.2023).

EDN
JHTACG

Bibliographic description
Fomin, I.S. (2024), "Comparison of compressed vector representations of matrices using compartment spiking neuron model (CSNM)", Robotics and Technical Cybernetics, vol. 13, no. 1, pp. 33-40, EDN: JHTACG. (in Russian).

UDC identifier
004.896

References

  1. Korsakov, A.M., Astapova, L.A. and Bakhshiev, A.V. (2022), “Application of a segmental spike neuron model with structural adaptation for solving classification problems”, Computer Science And Automation, 21, 3, pp. 493-520. (in Russian).
  2. Ciaparrone, G. et al. (2020), “Deep Learning in Video Multi-Object Tracking: A Survey”, Neurocomputing, 381, pp. 61-88, DOI: 10.1016/j.neucom.2019.11.023.
  3. Luo, W. et al. (2021), “Multiple Object Tracking: A Literature Review”, Artificial Intelligence, 293, 103448, DOI: 10.1016/j.artint.2020.103448.
  4. Bertinetto, L. et al. (2016), “Fully-Convolutional Siamese Networks for Object Tracking. Computer Vision”, in: ECCV 2016 Workshops : Lecture Notes in Computer Science, in Hua, G. and Jégou, H. ( ed.), Springer International Publishing, pp. 850-865, DOI: 10.1007/978-3-319-48881-3_56.
  5. Kim, M., Alletto, S. and Rigazio, L. (2017), “Similarity Mapping with Enhanced Siamese Network for Multi-Object Tracking”, arXiv:1609.09156 [cs], arXiv, DOI: 10.48550/arXiv.1609.09156.
  6. Wang, B. et al. (2016), “Joint Learning of Convolutional Neural Networks and Temporally Constrained Metrics for Tracklet Association”, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 1-8, DOI: 10.1109/CVPRW.2016.55.
  7. Zhang, S. et al. (2016), “Tracking Persons-of-Interest via Adaptive Discriminative Features”, Computer Vision – ECCV 2016 : Lecture Notes in Computer Science, in Leibe, B. et al. (ed.), Springer International Publishing, pp. 415-433, DOI: 10.1007/978-3-319-46454-1_26.
  8. Son, J., Baek, M., Cho, M. and Han, B. (2017), “Multi-Object Tracking With Quadruplet Convolutional Neural Networks”, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5620-5629, DOI: 10.1109/CVPR.2017.403.
  9. Zhu, J., et al. (2018), “Online Multi-Object Tracking with Dual Matching Attention Networks”, in: Proceedings of the European Conference on Computer Vision (ECCV), pp. 366-382, DOI: 10.1007/978-3-030-01228-1_23.
  10. Hermans, A., Beyer, L. and Leibe, B. (2017), “In Defense of the Triplet Loss for Person Re-Identification”, arXiv:1703.07737 [cs], arXiv, DOI: 10.48550/arXiv.1703.07737.
  11. Zhou, H. et al. (2019), “Deep Continuous Conditional Random Fields With Asymmetric Inter-Object Constraints for Online Multi-Object Tracking”, IEEE Transactions on Circuits and Systems for Video Technology, vol. 29, no. 4, pp. 1011-1022, DOI: 10.1109/TCSVT.2018.2825679.
  12. Chen, L., Ai, H., Zhuang, Z. and Shang, C. (2018), “Real-Time Multiple People Tracking with Deeply Learned Candidate Selection and Person Re-Identification”, 2018 IEEE International Conference on Multimedia and Expo (ICME), 1-6, DIO: 10.1109/ICME.2018.8486597.
  13. Dai, J., Li, Y., He, K. and Sun, J. (2016), “R-FCN: Object Detection via Region-based Fully Convolutional Networks”, Advances in Neural Information Processing Systems, Curran Associates, Inc., 29, DOI: 10.48550/arXiv.1605.06409.
  14. Bakhshiev, A., Demcheva, A. and Stankevich, L. (2022), “CSNM: The Compartmental Spiking Neuron Model for Developing Neuromorphic Information Processing Systems”, Advances in Neural Computation, Machine Learning, and Cognitive Research V : Studies in Computational Intelligence, in Kryzhanovsky, B. et al. (ed.), Springer International Publishing, pp. 327-333, DOI: 10.1007/978-3-030-91581-0_43.
  15. Leal-Taixé, L. et al. (2015), “Towards a benchmark for multi-target tracking”, arXiv preprint arXiv:1504.01942, DOI: 10.48550/arXiv.1504.01942
  16. Bakhshiev, A. and Gundelakh, F. (2015), “Mathematical Model of the Impulses Transformation Processes in Natural Neurons for Biologically Inspired Control Systems Development”, in: CEUR Workshop Proceedings, 1452, 1.
  17. Kaggle, “Triplet Loss with PyTorch”, available at: https://kaggle.com/code/hirotaka0122/triplet-loss-with-pytorch (Accessed 30 August 2023).
  18. He, K., Zhang, X., Ren, S. and Sun, J. (2016), “Deep Residual Learning for Image Recognition”, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770-778, DOI: 10.48550/arXiv.1512.03385.
  19. Bucher, M., Herbin, S. and Jurie, F. (2016), “Hard negative mining for metric learning based zero-shot classification”, in: Computer Vision–ECCV 2016 Workshops, Amsterdam, Netherlands, October 8-10 and 15-16, Springer International Publishing, pp. 524-531, DOI: 10.1007/978-3-319-49409-8_45.