Search

Batch normalization is one of the most widely used components in deep neural networks. It can accelerate training, and boost model performance on normal samples. However, batch normalization induces vulnerability to models on adversarial examples, especially in medical images, and the reason is still not clear. In this paper, we aim to explain the vulnerability aroused by batch normalization under adversarial images. Specifically, we first discover that both natural and medical images contain a large number of trivial features, whose weights will be enlarged under adversarial attacks, and batch normalization can further enlarge their weights. Additionally, we find that batch normalization will reduce the inter-class margin of high-level features, leading to less tolerance to adversarial perturbations, thereby decreasing the model robustness. Moreover, we hypothesize that the smaller inter-class margin, the more difficult to attain the optimal classification space, which means batch normalization will restrict the performance of adversarial training. This further verifies that a narrower inter-class margin induced by batch normalization reduces the model robustness. Experiments on four benchmark datasets demonstrate our discovery, interpretation and hypothesis. [ABSTRACT FROM AUTHOR]

In this paper, experimental and numerical study were both conducted to investigate the effect of pore size and porosity distribution on radiation absorption and thermal performance of porous solar energy absorber. Ultraviolet-visible-near infrared (UV-Vis-NIR) spectrophotometer was used to measure the transmittance of porous media to reflect its radiation absorption capabilities. Numerical model was established based on the assumption of thermal nonequilibrium condition as well as using P1 model to consider the radiation heat transfer. The UV-Vis-NIR spectrophotometer measurement showed that: (1) With smaller pore size, the spectral transmittance of the porous media would be lower and the solar radiation absorption would be better; (2) Among the materials with different pore size distributions, pore-size-decreased combo and pore-size-increased combo have almost equal absorption coefficient which are higher than that of uniform structure. Numerical simulation demonstrated that: (3) For materials with different pore size distributions, pore-size-decreased structure has the best radiation absorption due to its ability of maximizing the volumetric absorption effect, which is agreed with the UV-Vis-NIR spectrophotometer experimental results; (4) For materials with different porosity distributions, porosity-gradually-increased structure has the highest mean fluid/solid temperatures because it can utilize the enhanced convective/conductive heat transfer to improve the overall thermal performance of porous receiver; (5) Porous structure with pore-size-decreased distribution and porosity-gradually-increased distribution has the best thermal performance of which the mean temperatures of fluid/solid phases are the highest among all the studied cases. [ABSTRACT FROM AUTHOR]