A Monte Carlo Method for Image Classification Using SVM
DOI:
https://doi.org/10.55630/dipp.2021.11.20Keywords:
Deep Neural Networks, Monte Carlo method, Image Classification, Support Vector MachinesAbstract
Support Vector Machines are a widely used tool in Machine Learning. They have some important advantages with regards to the more popular Deep Neural Networks. For the problem of image classification, multiple SVMs may be used and the issue of finding the best hyperparameters adds additional complexity and increases the overall computational time required. Our goal is to develop and study Monte Carlo algorithms that allow faster discovery of good hyperparameters and training of the SVMs, without impacting negatively the final accuracy of the models. We also employ GPUs and parallel computing in order to achieve good utilisation of the capabilities of the available hardware. In this paper we describe our methods, provide implementation details and show numerical results, achieved on the publicly available Architectural Heritage Elements image Dataset.References
Abed, M. A.-A. (2020). Architectural heritage images classification using deep learning. Proceedings of the 2nd International Workshop on Visual Pattern.
Alexandrov, A. K. (2005). Finding the Smallest Eigenvalue by the Inverse Monte Carlo Method with Refinement. ICCS 2005: Computational Science – ICCS 2005. , pp. 766-774. Lecture Notes in Computer Science. doi:doi.org/10.1007/11428862_104
Boser, B. E. (1992). A Training Algorithm for Optimal Margin Classifiers. (COLT'92), Proceedings of the 5th Annual Workshop on Computational Learning Theory. Pittsburgh, PA: USA: ACM Press. doi:doi.org/10.1145/130385.130401
Claesen, M. &. (2015). Hyperparameter Search in Machine Learning. MIC 2015: The XI Metaheuristics International Conference, CoRR, abs/1502.02127. .
Cristianini, N. S.-T. (2020). An Introduction to Support Vector Machines and Other Kernel-based Learning Methods. Cambridge University Press, ISBN: 0521780195.
Dimov, I. D. (1999). A new iterative Monte Carlo approach for inverse matrix problem. Journal of Computational and Applied Mathematic, 92 (1), 15-35.
Fraccaroli, M. L. (2020). Automatic Setting of DNN Hyper-Parameters by Mixing Bayesian Optimization and Tuning Rules. (O. a. Nicosia G. et al. (eds) Machine Learning, Ed.) Lecture Notes in Computer Science, 12565 . doi:doi.org/10.1007/978- 3-030-64583-0_43
Kroese, D. B. (2019). Data Science and Machine Learning: Mathematical and Statistical Methods. Chapman and Hall/CRC.
Llamas, J. L.-G.-B. (2017). Classification of Architectural Heritage Images Using Deep Learning Techniques. Applied Sciences , 992.
Mikhailov, G. (1992). Optimization of Weighted Monte Carlo Methods. Springer- Verlag Berlin-Heidelberg.
NVIDIA . (2017). Retrieved from NVIDIA Tesla V100 GPU Architecture the world’s Most Advanced Data Center GPU. White paper http://www.nvidia.com/object/volta-architecture-whitepaper.html.
Rudi, A. C. (2017). FALKON: an optimal large scale kernel method. NIPS'17: Proceedings of the 31st International Conference on Neural Information Processing Systems , (pp. 3891–3901).
Snoek, J. L. (2012). Practical Bayesian Optimization of Machine Learning Algorithms. Advances in Neural Information Processing Systems , , pp. 2960-2968.
