While neural models show remarkable accuracy on individual predictions, their internal beliefs can be inconsistent across examples. In this paper, we formalize such inconsistency as a generalization of prediction error. We propose a learning framework for constraining models using logic rules to regularize them away from inconsistency. Our framework can leverage both labeled and unlabeled examples and is directly compatible with off-the-shelf learning schemes without model redesign. We instantiate our framework on natural language inference, where experiments show that enforcing invariants stated in logic can help make the predictions of neural models both accurate and consistent.

In recent years, experts and scholars in the field of information security have attached great importance to the security of image information. They have proposed many image encryption algorithms with higher security. In order to further improve the security level of image encryption algorithm, this paper proposes a new image encryption algorithm based on two-dimensional Lorenz and Logistic. The encryption test of several classic images proves that the algorithm has high security and strong robustness. This paper also analyzes the security of encryption algorithms, such as analysis of the histogram, entropy process of information, examination of correlation, differential attack, key sensitivity test, secret key space analysis, noise attacks, contrast analysis. By comparing the image encryption algorithm proposed in this paper with some existing image encryption algorithms, the encryption algorithm has the characteristics of large secret key space, sensitivity to the key, small correlation coefficient and high contrast. In addition, the encryption algorithm is used. It can also resist noise attacks.

With the recent advances in deep learning, neural network models have obtained state-of-the-art performances for many linguistic tasks in natural language processing. However, this rapid progress also brings enormous challenges. The opaque nature of a neural network model leads to hard-to-debug-systems and difficult-to-interpret mechanisms. Here, we introduce a visualization system that, through a tight yet flexible integration between visualization elements and the underlying model, allows a user to interrogate the model by perturbing the input, internal state, and prediction while observing changes in other parts of the pipeline. We use the natural language inference problem as an example to illustrate how a perturbation-driven paradigm can help domain experts assess the potential limitation of a model, probe its inner states, and interpret and form hypotheses about fundamental model mechanisms such as attention.

No abstract

As one of the largest coal-rich provinces in China, Shanxi has extensive underground coal-mining operations. These operations have caused numerous ground cracks and substantial environmental damage. To study the main geological and mining factors influencing mining-related ground cracks in Shanxi, a detailed investigation was conducted on 13 mining-induced surface cracks in Shanxi. Based on the results, the degrees of damage at the study sites were empirically classified into serious, moderate, and minor, and the influential geological and mining factors (e.g., proportions of loess and sandstone in the mining depth, ratio of rock thickness to mining thickness, and ground slope) were discussed. According to the analysis results, three factors (proportion of loess, ratio of rock thickness to mining thickness, and ground slope) play a decisive role in ground cracks and can be respectively considered as the critical material, mechanical, and geometric conditions for the occurrence of mining surface disasters. Together, these three factors have a strong influence on the occurrence of serious discontinuous ground deformation. The results can be applied to help prevent and control ground damage caused by coal mining. The findings also provide a direct reference for predicting and eliminating hidden ground hazards in mining areas.