Steganalysis of neural networks based on parameter statistical bias

Many pretrained deep learning models have been released to help engineers and researchers develop deep learning-based systems or conduct research with minimall effort. Previous work has shown that at secret message can be embedded in neural network parameters without compromising the accuracy of the model. Malicious developers can, therefore, hide malware or other baneful information in pretrained models, causing harm to society. Hence, reliable detection of these vicious pretrained models is urgently needed. We analyze existing approaches for hiding messages and find that they will ineluctably cause biases in the parameter statistics. Therefore, we propose steganalysis methods for steganography on neural network parameters that extract statistics from benign and malicious models and build classifiers based on the extracted statistics. To the best of our knowledge, this is the first study on neural network steganalysis. The experimental results reveal that our proposed algorithm can effectively detect a model with an embedded message. Notably, our detection methods are still valid in cases where the payload of the stego model is low.     

基于无人机实时数据多波次任务规划模型分析

针对导弹部队多波次作战任务规划问题, 依据无人机的实时数据, 构建了基于路径的多层规划模型, 并设计了模型的算法求解流程。使用遗传算法与禁忌搜索混合算法, 得出了任务规划中的最优路径规划, 并在此基础上进行了冲突的消除。通过仿真案例表明, 用无人机协同配合导弹部队作战, 实时传输作战数据, 能够解决战场信息模糊不确定的问题; 使用多层规划模型能够为导弹多波次规划作战的路径进行科学的决策和选择。利用遗传算法和禁忌搜索混合算法, 能够避免局部最优导致无法输出结果的现象。     

The Effectiveness of Structural Monetary Policy and Macro-Prudential Policies — Based on the DSGE Model That Includes Bank Heterogeneous Credit

Journal of Systems Science and Complexity - Structural monetary policy and macro-prudential policies are important parts of the policy system of the People’s Bank of China. By constructing a...     

基于改进Faster R-CNN的农田残膜识别方法

为了实现农田残膜的精准捡拾,提高残膜回收机的回收率.将改进Faster R-CNN卷积神经网络运用到农田残膜的识别检测中,提出了一种农田残膜的识别方法.以11MS-1850残膜回收机工作后遗留在农田表面的残膜为研究对象,分别在晴天、阴天不同时间段采集图像共计1648幅.通过更改图像亮度、旋转等方式扩充数据集,最终得到4950幅残膜样本图像,按照7:2:1划分为训练集(3465幅)、验证集(990幅)、测试集(495幅);采用双阈值算法替代传统的单阈值算法,降低了阈值对模型性能的影响;通过对比试验,选取具有残差网络结构的ResNet50作为主干特征提取网络,准确率可达88.84％,召回率为87.70％,总体精度为88.27％;为了使检测模型对小目标更加灵敏,根据数据集中残膜尺寸大小,在原有锚点基础上增加322和642的尺度参数,准确率、召回率、总体精度分别提升了1.29％、0.67％、0.97％,单幅检测时间为284.13 ms,基本满足了识别残膜的要求.可为残膜回收机加装补收装置提供参考,为研制人工智能残膜回收机提供理论基础.     

多维度深海鱼类识别算法

针对深海光线分布不均匀导致鱼类识别检测困难的问题,提出了符合视觉认知的多维度深海鱼类识别算法.该方法从时间维度优化传统的高斯混合模型(GMM)初步确定变化区域,从空间维度构建目标特征,完整提取运动目标,从时空关联维度建立深度学习的鱼类识别框架,试验结果表明:本算法可在多种复杂条件下准确提取运动目标,面积交迭度(AOM)达到80％以上,优于当前主流算法.     

化学实验教学创新设计的调查报告

为了了解化学实验教学创新设计的现状,对常德市普通高中化学教师进行问卷调查,调查结果显示高中化学实验教学创新设计有弱化的趋势.在此基础上,从学科教学的角度对弱化的原因进行了分析,并提出化学实验教学创新设计的OCPP评价模式,尽可能建构、揭示或阐述实验教学创新设计的基本原理.OCPP模式包括目标评价(Objective Evaluation)、内容评价(Content Evaluation)、过程评价(Process Evaluation)和成果评价(Product Evaluation)4个部分.     

科技成果转化的探索与思考 ——基于湖南省科技成果转化的实证研究

科技成果转化是推动科技与经济结合的直接动力,是促进经济社会高质量发展的必要途径.文章通过对科技成果转化的重要性、现状和障碍的分析,提出科技行政管理部门应采取的管理举措.     

考虑多个子群意见的多要素大群体决策方法

针对存在多个子群的多要素大群体决策问题，提出一种决策分析方法.首先，依据参与决策的个体针对方案构成要素的排序，计算各要素针对各参与个体的borda分值.其次，构建要素针对各子群的平均borda分值向量，再将其归一化后作为针对各子群的虚拟方案.然后，考虑到各子群的评估一致性存在差异，计算各子群的权重.进一步地，计算各虚拟方案与各备选方案之间的贴近度，并将其与各子群权重相集结，得到针对各备选方案的综合贴近度，并据此对备选方案进行排序.最后，通过一个考虑公众意愿的城市公园设计方案选择问题的算例说明了所提出方法的可用性.     

Blockage of the NLRP3 inflammasome by MCC950 improves anti-tumor immune responses in head and neck squamous cell carcinoma

The NLRP3 inflammasome is a critical innate immune pathway responsible for producing active interleukin (IL)-1β, which is associated with tumor development and immunity. However, the mechanisms regulating the inflammatory microenvironment, tumorigenesis and tumor immunity are unclear. Herein, we show that the NLRP3 inflammasome was over-expressed in human HNSCC tissues and that the IL-1β concentration was increased in the peripheral blood of HNSCC patients. Additionally, elevated NLRP3 inflammasome levels were detected in tumor tissues of Tgfbr1/Pten 2cKO HNSCC mice, and elevated IL-1β levels were detected in the peripheral blood serum, spleen, draining lymph nodes and tumor tissues. Blocking NLRP3 inflammasome activation using MCC950 remarkably reduced IL-1β production in an HNSCC mouse model and reduced the numbers of myeloid-derived suppressor cells (MDSCs), regulatory T cells (Tregs) and tumor-associated macrophages (TAMs). Moreover, inhibiting NLRP3 inflammasome activation increased the numbers of CD4⁺ and CD8⁺ T cells in HNSCC mice. Notably, the numbers of exhausted PD-1⁺ and Tim3⁺ T cells were significantly reduced. A human HNSCC tissue microarray showed that NLRP3 inflammasome expression was correlated with the expression of CD8 and CD4, the Treg marker Foxp3, the MDSC markers CD11b and CD33, and the TAM markers CD68 and CD163, PD-1 and Tim3. Overall, our results demonstrate that the NLRP3 inflammasome/IL-1β pathway promotes tumorigenesis in HNSCC and inactivation of this pathway delays tumor growth, accompanied by decreased immunosuppressive cell accumulation and an increased number of effector T cells. Thus, inhibition of the tumor microenvironment through the NLRP3 inflammasome/IL-1β pathway may provide a novel approach for HNSCC therapy.