Performance Analysis of Distributed Deep Learning using Horovod for Image Classification
R M Rakshith, Vineet Lokur, Prateek Hongal, Vivek Janamatti, Satyadhyan Chickerur · 2022 6th International Conference on Intelligent Computing and Control Systems (ICICCS) · 2022
Recent advances in unsupervised feature learning and deep learning methodologies have shown that training large models may significantly improve performance. This research study reviews the topic of training a deep neural network in a distributed manner. The proposed study uses the Horovod to train large models on compute clusters with thousands of computers. It can be viewed that the image classification and segmentation are computationally expensive jobs, with more segmentation. Lower-level tasks may be fine-tuned for enhanced efficiency, but the study demonstrates that the proposed research study can work better and more effectively in categorization by exploiting the distributed deep learning idea. The proposed study demonstrate the application of Horovod on Image classification using CNNs, and our works serve as a benchmark for performance analysis on standard datasets; CIFAR 10 and Cat Vs Dog.