Ultralight ceramic aerogels are attractive thermal superinsulating materials, but display a formidable tradeoff between low and high temperature thermal conductivity (κ) due to their low-density features. Embedding carbon species as infrared opacifier in ultralight ceramic aerogels can substantially reduce the thermal radiation heat transfer without compromising the ultralow solid conduction. However, the oxidation resistance of embedded carbon species still remains inadequate to prevent thermal etching at high temperatures. Herein, we report a carbonaceous design and synthesis of ceramic nanofibrous aerogels with amorphous carbon embedded in the yttrium-stabilized zircon nanofibers to achieve a high-temperature thermal superinsulating performance with robust thermomechanical stability. The aerogels display one of the lowest κ of 95 mW·m⁻¹·K⁻¹ at 1,000 °C in air among ultralight material family, as well as robust mechanical flexibility with up to 95% compressive strain, 30% non-linear fracture strain, and 99% bending strain, and high thermal stability with ultralow strength degradation less than 1% after sharp thermal shocks (240 °C·s⁻¹) and working temperature up to 1,200 °C. The combined high-temperature thermal superinsulating and thermomechanical properties offer an attractive material system for robust thermal insulation under extreme conditions.

Generative adversarial networks (GANs) are an unsupervised generative model that learns data distribution through adversarial training. However, recent experiments indicated that GANs are difficult to train due to the requirement of optimization in the high dimensional parameter space and the zero gradient problem. In this work, we propose a self-sparse generative adversarial network (Self-Sparse GAN) that reduces the parameter space and alleviates the zero gradient problem. In the Self-Sparse GAN, we design a self-adaptive sparse transform module (SASTM) comprising the sparsity decomposition and feature-map recombination, which can be applied on multi-channel feature maps to obtain sparse feature maps. The key idea of Self-Sparse GAN is to add the SASTM following every deconvolution layer in the generator, which can adaptively reduce the parameter space by utilizing the sparsity in multi-channel feature maps. We theoretically prove that the SASTM can not only reduce the search space of the convolution kernel weight of the generator but also alleviate the zero gradient problem by maintaining meaningful features in the batch normalization layer and driving the weight of deconvolution layers away from being negative. The experimental results show that our method achieves the best Fréchet inception distance (FID) scores for image generation compared with Wasserstein GAN with gradient penalty (WGAN-GP) on MNIST, Fashion-MNIST, CIFAR-10, STL-10, mini-ImageNet, CELEBA-HQ, and LSUN bedrooms datasets, and the relative decrease of FID is 4.76%–21.84%. Meanwhile, an architectural sketch dataset (Sketch) is also used to validate the superiority of the proposed method.