Synthetic Dataset for Deep Learning

Synthetic Dataset for Deep Learning是由特拉华大学电气与计算机科学系创建的一个合成数据集，旨在为深度学习提供一个具有明确高斯分布的实验工具。该数据集包含70,000张32×32的灰度图像，分为10个类别，每个类别有6,000张训练图像和1,000张测试图像，与MNIST数据集具有相同的特性。数据集的创建过程涉及从NIST数据集中提取图像，并通过特定的算法将其转换为服从高斯分布的合成图像。该数据集主要用于验证深度学习理论，特别是深度神经网络的层次性和充分性等特性。

The Synthetic Dataset for Deep Learning is a synthetic dataset developed by the Department of Electrical and Computer Engineering at the University of Delaware. It is designed as an experimental tool with a well-defined Gaussian distribution for deep learning research. This dataset contains 70,000 32×32 grayscale images, divided into 10 categories, with 6,000 training images and 1,000 test images per category, sharing the same characteristics as the MNIST dataset. The dataset creation process involves extracting images from the NIST dataset and converting them into synthetic images that follow a Gaussian distribution using a dedicated algorithm. This dataset is primarily used to validate deep learning theories, especially properties including the hierarchical structure and sufficiency of deep neural networks.