DeepNets-1M

该数据集名为DeepNets-1M，包含以图形方式表示的神经网络架构，总量达到100万个。在这100万个架构中，仅有1,000个在CIFAR-10数据集上进行了评估并获得了性能标签。具体来说，这个数据集中有500个架构用于训练，另外500个用于测试，其性能标签基于在CIFAR-10数据集上的准确度、推理时间和收敛时间。规模上，该数据集包含100万个神经网络架构，其中1,000个带有性能标签。其任务是预测神经网络的性能。

This dataset, named DeepNets-1M, consists of 1,000,000 graphically represented neural network architectures. Out of these one million architectures, only 1,000 have been evaluated on the CIFAR-10 dataset and assigned performance labels. Specifically, 500 of the labeled architectures are used for training, while the remaining 500 are reserved for testing. The performance labels are based on the classification accuracy, inference time, and convergence time on the CIFAR-10 dataset. In terms of scale, the dataset contains 1 million neural network architectures, with only 1,000 of them carrying performance labels. The core task of this dataset is to predict the performance of neural networks.