Improving Resnet-9 Generalization Trained on Small Datasets
This work addresses the challenge of efficient training for hardware-aware applications, but it is incremental as it combines existing techniques for a competition setting.
The paper tackled the problem of achieving high accuracy in image classification with limited training time and data, specifically training ResNet-9 on a small subset of CIFAR-10 in under 10 minutes, resulting in an accuracy of 88% on a secret evaluation dataset.
This paper presents our proposed approach that won the first prize at the ICLR competition on Hardware Aware Efficient Training. The challenge is to achieve the highest possible accuracy in an image classification task in less than 10 minutes. The training is done on a small dataset of 5000 images picked randomly from CIFAR-10 dataset. The evaluation is performed by the competition organizers on a secret dataset with 1000 images of the same size. Our approach includes applying a series of technique for improving the generalization of ResNet-9 including: sharpness aware optimization, label smoothing, gradient centralization, input patch whitening as well as metalearning based training. Our experiments show that the ResNet-9 can achieve the accuracy of 88% while trained only on a 10% subset of CIFAR-10 dataset in less than 10 minuets