基于1D CNN与XGBoost的恶意代码纹理检测

恶意代码数量已经呈现爆炸式增长,对于恶意代码的检测防护显得尤为重要.近几年,基于深度学习的恶意代码检测方法开始出现,基于此,提出一种新的检测方法,将恶意代码二进制文件转化为十进制数组,并利用一维卷积神经网络(1 Dimention Convolutional Neural Networks,1D CNN)对数组进行分类和识别.针对代码家族之间数量不平衡的现象,该算法选择在分类预测上表现良好的XGBoost,并对Vision Research Lab中的25个不同恶意软件家族的9458个恶意软件样本进行了实验.实验结果表明,所提的方法分类预测精度达到了97％.     

基于GA_XGBoost模型的大学生科研能力预测问题研究

科学评价大学生科研创新能力对我国科研水平的提高具有重要意义.采用机器学习模型来预测大学生科研能力可以起到良好的效果,提出一种GAXGBoost模型来实现对大学生的科研能力预测.此模型是以Xgboost算法为基础,然后充分利用遗传算法的全局搜索能力自动搜索Xgboost最优超参数,避免了人为经验调参不准确的缺陷,最后采用精英选择策略以此确保每一轮都是最佳的进化结果.通过分析表明,所采用的GAXGBoost模型在大学生科研能力预测的结果中具有很高的精度,将此模型与Logistic Regression、Random Forest、SVM等模型进行对比,GAXGBoost模型的预测精度最高.     

基于图像分割技术的图像数字水印算法研究

针对数字图像水印技术,通过对闭运算得到的图像再进行开运算,使实验得到的图像区域较为完整,采用4×4的正方形结构元素,使得图像区域有更多的白色像素被置为黑色,对图像进行取反操作,能够对图像区域的非连通区域进行填充.希望能够利用精确的图像分割算法,实现对复杂版面的文档图像的有效切割,在一定程度上对图像数字水印的推广起到促进作用.     

Single Crystal to Single Crystal (SC‐to‐SC) Transformation from a Nonporous to Porous Metal–Organic Framework and Its Application Potential in Gas Adsorption and Suzuki Coupling Reaction through Postmodification

A new amino‐functionalized strontium–carboxylate‐based metal–organic framework (MOF) has been synthesized that undergoes single crystal to single crystal (SC‐to‐SC) transformation upon desolvation. Both structures have been characterized by single‐crystal X‐ray analysis. The desolvated structure shows an interesting 3D porous structure with pendent ?NH₂ groups inside the pore wall, whereas the solvated compound possesses a nonporous structure with DMF molecules on the metal centers. The amino group was postmodified through Schiff base condensation by pyridine‐2‐carboxaldehyde and palladium was anchored on that site. The modified framework has been utilized for the Suzuki cross‐coupling reaction. The compound shows high activity towards the C?C cross‐coupling reaction with good yields and turnover frequencies. Gas adsorption studies showed that the desolvated compound had permanent porosity and was microporous in nature with a BET surface area of 2052 m² g^?1. The material also possesses good CO₂ (8 wt %) and H₂ (1.87 wt %) adsorption capabilities.     

基于LED的多色温多星等单星模拟器

针对星模拟器与星敏感器观星的色温不匹配对星敏感器光信号定标精度产生的影响,设计了一种基于LED的多色温多星等单星模拟器,采用该系统模拟特定色温的星光用于星敏感器光信号定标,大幅降低了色温非匹配带来的定标误差。从理论上分析了色温非匹配影响星敏感器光信号的定标精度的机理;根据设计指标确定了星模拟器的设计方案,主要解决了光源的选型、多色温多星等单星模拟器驱动和控制系统、色温星等匹配算法、多色温多星等单星模拟器软件设计四项关键技术问题;对多色温多星等单星模拟器进行了标定和性能测试:0等星4000K和3等星7000 K星光的光谱匹配误差分别为4.87%和7.83%,星等等级分别为0.03和2.93;光源稳定后,多色温多星等单星模拟器的平行光管出口Φ100 mm口径内的照度非均匀度为6.5%,均满足设计指标。     

Polypyrrole Shell@3D‐Ni Metal Core Structured Electrodes for High‐Performance Supercapacitors

Three‐dimensional (3D) nanometal films serving as current collectors have attracted much interest recently owing to their promising application in high‐performance supercapacitors. In the process of the electrochemical reaction, the 3D structure can provide a short diffusion path for fast ion transport, and the highly conductive nanometal may serve as a backbone for facile electron transfer. In this work, a novel polypyrrole (PPy) shell@3D‐Ni‐core composite is developed to enhance the electrochemical performance of conventional PPy. With the introduction of a Ni metal core, the as‐prepared material exhibits a high specific capacitance (726 F g^?1 at a charge/discharge rate of 1 A g^?1), good rate capability (a decay of 33 % in C_sp with charge/discharge rates increasing from 1 to 20 A g^?1), and high cycle stability (only a small decrease of 4.2 % in C_sp after 1000 cycles at a scan rate of 100 mV s^?1). Furthermore, an aqueous symmetric supercapacitor device is fabricated by using the as‐prepared composite as electrodes; the device demonstrates a high energy density (≈21.2 Wh kg^?1) and superior long‐term cycle ability (only 4.4 % and 18.6 % loss in C_sp after 2000 and 5000 cycles, respectively).     

On Minimal Asymptotic Bases

Mathematical Notes - Let $$\mathbb N$$ denote the set of all nonnegative integers, and let $$A\subseteq\mathbb N$$ . Let $$h,n\in\mathbb N$$ , $$h\ge 2$$ and $$r_h(A,n)=\#\{(a_1,\dots,a_h)\in...     

紫色球杆菌视紫红质光谱特性的机器学习研究

近年来，机器学习等人工智能技术被应用于蛋白质工程，其在蛋白质结构、功能预测、催化活性等研究中具有独特优势。在未知蛋白质结构的情况下，将蛋白质序列和功能特性与机器学习相结合，基于序列-活性关系（innovative sequence-activity relationship，ISAR）算法，将蛋白质氨基酸序列数字化，用快速傅里叶变换（fast four transform，FFT）进行预处理，再进行偏最小二乘回归建模，可在数据集较少情况下拟合得到最佳模型。通过机器学习对紫色球杆菌视紫红质（gloeobacter violaceus rhodopsin，GR）的突变体蛋白质氨基酸序列与光谱最大吸收波长进行建模，获得了最佳模型。用最佳索引LEVM760106建模得到的确定系数R² 为0.944，均方误差E为11.64。用小波变换进行的预处理，其R² 虽也约为0.944，但E大于11.64，不及FFT进行的预处理。方法较好地解决了蛋白质序列与功能特性之间的数学建模问题，在蛋白质工程中可为预测更优的突变体提供支持。