Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{186590,
        author = {Kalpana Dongre and Bhakti Radheshyam Thombare},
        title = {A Hybrid Optimization Framework to Reduce Deep Learning Training Time Using Mixed Precision, Gradient Accumulation, and Adaptive Batch Scaling},
        journal = {International Journal of Innovative Research in Technology},
        year = {2025},
        volume = {12},
        number = {6},
        pages = {2078-2080},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=186590},
        abstract = {Deep learning models have achieved state-of-the-art performance in several domains, but their training time remains a major bottleneck due to high computational cost, large-scale datasets, and increasingly complex architectures. This paper proposes a hybrid optimization framework combining mixed precision training, adaptive batch scaling, and gradient accumulation to significantly reduce model training time without compromising accuracy. Experimental validation on ResNet-50 and Transformer models using the CIFAR-100 and IMDB sentiment datasets shows up to 38.6% reduction in total training time and 22.4% lower GPU memory usage while maintaining accuracy within ±0.5% of full-precision baselines. The study integrates insights from state-of-the-art works on efficient DL training, but introduces a unified workflow capable of running on a single GPU, multi-GPU, or cloud compute environment. The proposed approach provides a practical and scalable solution for researchers and industries aiming to reduce computational cost in model development.},
        keywords = {Deep learning optimization, mixed precision, batch scaling, training acceleration, GPU efficiency, model training time reduction.},
        month = {November},
        }