ON THE FEATURES OF THE FORMATION OF INPUT DATA IN THE SIAMESE NEURAL NETWORK
DOI:
https://doi.org/10.26906/SUNZ.2024.3.193Keywords:
Siamese neural network, descriptor, neural network testingAbstract
Various methods of forming input data and descriptor estimates of the Siamese neural network (SNM) for comparing images of handwritten digits are analyzed. A method of using a quasi-random N-dimensional sequence of vectors, formed according to Sobol's method, is proposed for the formation of descriptors, which, together with images, are input data for training SNM and its further use. Siamese neural network testing was performed using the received evaluations of image descriptors of handwritten digits. The MNIST set was used during SNM testing. The result of testing the SNM model gave a value of the accuracy indicator equal to 0.9706. The test results showed that the considered estimates of the descriptors h1 and h2 reduce the number of errors during testing compared to the use of the descriptor h0 based on a quasi-random N-dimensional sequence of vectors.Downloads
References
Chicco D. Siamese Neural Networks: An Overview. Artificial Neural Networks. 2021. MIMB, vol. 2190. P. 73–94. URL: https://link.springer.com/protocol/10.1007/978-1-0716-0826-5_3
Шостак А. В. Про особливості формування дескрипторів у сіамській нейронній мережі. Системи управління, навігації та зв'язку. Полтава: НУ ПП, 2021. Вип. 4(66). С. 91–96. DOI: https://doi.org/10.26906/SUNZ.2021.4.079
Contrastive loss for Siamese networks with Keras and TensorFlow. URL: https://www.pyimagesearch.com/2021/01/18/contrastive-loss-for-siamese-networks-with-keras-and-tensorflow/
Image similarity estimation using a Siamese Network with a contrastive loss. URL: https://keras.io/examples/vision/siamese_contrastive/
The Mnist database of handwritten digits. URL: http://yann.lecun.com/exdb/mnist/
Owen, A.B. On Dropping the First Sobol’ Point. In: Keller, A. (eds) Monte Carlo and Quasi-Monte Carlo Methods. MCQMC 2020. Springer Proc. in Mathematics & Statistics, vol 387. Springer, Cham. DOI: https://doi.org/10.1007/978-3-030-98319-2_4
Roberts M. The Unreasonable Effectiveness of Quasirandom Sequences. 2020. URL: https://extremelearning.com.au/unreasonable-effectiveness-of-quasirandom-sequences/
Halton J. H. On the efficiency of certain quasi-random sequences of points in evaluating multidimensional integrals. Numer. Math. 1960. Vol. 2. P. 84–90. DOI: https://doi.org/10.1007/BF01386213
