Rapid-INR: Accelerating Deep Neural Network Training by Compressing Datasets Using Implicit Neural Representations
Rapid-INR utilizes Implicit Neural Representations (INR) to compress entire image datasets, enabling end-to-end training on GPU without the need for external memory access or powerful CPUs. This approach significantly accelerates training while maintaining high accuracy.