Autors: Slavov, D. V., Hristov, V. D., Slavova, A. V.
Title: Distributed Machine Learning through Transceiver Competitive Connectivity of Remote Computing Systems
Keywords: machine learning, distributed algorithms, PyTorch, data para

Abstract: An architecture is built for sequential data transmission among individual machines participating in a decentralized distributed machine learning of a convolutional neural network for image classification. Simultaneously running multiple replicas of a machine learning task using TCP communication protocol has been successfully implemented. A unification of the final result of each epoch until the end of the training session is performed on each of the participating machines in the training process. A working image classification model is obtained on all machines involved in the process. The results from the experiments performed herein show a 63% improvement in training speed for the fastest configuration with only a 3% drop in accuracy compared to the most accurate one. The study therefore confirms that the use of distributed learning systems can have significant benefits for the AI researchers and practitioners.

References

    Issue

    2023 International Scientific Conference on Computer Science (COMSCI), pp. 1-7, 2023, Bulgaria, DOI 10.1109/COMSCI59259.2023.10315948

    Вид: публикация в международен форум, публикация в реферирано издание, индексирана в Scopus