SYSTEMATIC ANALYTIC-GEOMETRIC AUGMENTATION FOR ROBUST DEEP LEARNING
Robustness to geometric distortions and distribution shifts remains a major challenge in deep learning. Standard data augmentation methods, such as random rotations and translations, improve generalization but often fail to systematically cover the geometric transformation space. In this work, we propose an analytic-geometric data augmentation framework that employs controlled affine transformations—including fixed-angle rotations, shearing, and scaling—defined analytically for each training sample. We hypothesize that such systematic augmentation better exposes neural networks to the full spectrum of geometric variations, enhancing their ability to generalize and remain robust under previously unseen conditions. We conduct extensive experiments on MNIST, FashionMNIST, and CIFAR-10 (grayscale), training lightweight convolutional models with both standard and analytic-geometric augmentations. Evaluation on clean, noisy, and rotated test sets demonstrates that analytic-geometric augmentation consistently yields higher accuracy and significantly improved robustness across all scenarios. Notably, our method boosts performance under noise and out-of-distribution rotations by up to 30% compared to conventional augmentations. Our findings highlight the value of explicitly structured augmentation for building more robust and reliable computer vision systems, and suggest that analytic-geometric methods should be incorporated into standard deep learning pipelines for applications where robustness is critical.
[1]. Simard, P. Y., Steinkraus, D., & Platt, J. C. (2003). Best practices for convolutional neural networks applied to visual document analysis. In ICDAR.
[2]. LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278-2324.
[3]. Shorten, C., & Khoshgoftaar, T. M. (2019). A survey on image data augmentation for deep learning. Journal of Big Data, 6(1), 60.
[4]. Goodfellow, I., Vinyals, O., & Saxe, A. (2015). Qualitatively characterizing neural network optimization problems. In ICLR.
[5]. Li, H., Xu, Z., Taylor, G., Studer, C., & Goldstein, T. (2018). Visualizing the loss landscape of neural nets. In NeurIPS.
[6]. Keskar, N. S., Mudigere, D., Nocedal, J., Smelyanskiy, M., & Tang, P. T. P. (2017). On large-batch training for deep learning: Generalization gap and sharp minima. In ICLR.
[7]. Dauphin, Y. N., Pascanu, R., Gulcehre, C., Cho, K., Ganguli, S., & Bengio, Y. (2014). Identifying and attacking the saddle point problem in high-dimensional non-convex optimization. In NeurIPS.
[8]. Taylor, G. W., & Nitschke, G. (2010). Improving deep learning using generic data augmentation. arXiv preprint arXiv:1708.06020.
[9]. Cubuk, E. D., Zoph, B., Mane, D., Vasudevan, V., & Le, Q. V. (2019). AutoAugment: Learning augmentation policies from data. In CVPR.
[10]. Hernandez-Garcia, A., & König, P. (2018). Data augmentation for deep learning: A review. arXiv preprint arXiv:1806.06898.
[11]. Hendrycks, D., & Dietterich, T. (2019). Benchmarking neural network robustness to common corruptions and perturbations. In ICLR.
[12]. Russakovsky, O., Deng, J., Su, H., et al. (2015). ImageNet Large Scale Visual Recognition Challenge. International Journal of Computer Vision, 115(3), 211–252.
[13]. Kingma, D. P., & Ba, J. (2015). Adam: A method for stochastic optimization. In ICLR.
[14]. Martens, J., & Grosse, R. (2015). Optimizing neural networks with Kronecker-factored approximate curvature. In ICLR.