打水漂与物理学规律

打水漂是人类最古老的游戏之一.通过对打水漂进行简化定量分析得出,石块弹跳,只有在抛掷的初速度超过某个临界值时才会发生,并且给出了石块的最大弹跳次数.     

Some fractal properties of the percolating backbone in two dimensions

A new algorithm is presented, based on elements of artificial intelligence theory, to determine the fractal properties of the backbone of the incipient infinite cluster. It is found that the fractal dimensionality of the backbone isd _f ^BB =1.61±0.01, the chemical dimensionality isd _t=1.40±0.01, and the fractal dimension of the minimum pathd _min=1.15 ± 0.02 for the two-dimensional triangular lattice.     

不需加人工变量的两阶段法

A method is provided to achieve an initial basic feasible solution of a linear programming in this paper. This method dose not need introducing any artificial variable, but needs only solving an auxiliary linear programming. Compared with the traditional two-phase method, it has advantages of saving the memories and reducing the computational efforts.     

GA-BP嵌套算法的理论及应用

分析了BP算法、遗传算法以及GA-BP-APARTING算法的特点,提出了GA-BP-NESTING算法.在人工神经网络的在线学习和离线学习方式下,分别对BP算法、GA算法、GA-BP-APARTING算法和GA-BP-NESTING算法进行了比较研究,研究发现:第一,网络初始权值的赋值对人工神经网络训练影响很大;第二,离线学习方式下GA-BP-NESTING算法效果最佳.     

一种新的拓扑指数用于手性化合物的构效关系研究

为了研究羟酸类和氨基酸类手性化合物的构效关系,本实验引进了手性乘积算法:P＝(1-b/a)(1-c/b)(1-a/c)(1-a/d),其中a、b、c和d分别为与分子中手性中心碳原子4个不同连接片断手性eAm指数和分子连接性指数mx的计算值,并考察了所得参数与羟酸类和氨基酸类化合物手性薄层色谱保留指数的相关性.为了评定所得预测模型的稳定性,文中运用了交叉验证法.结果表明,由人工神经网络方法(BP法)所得数学模型比较稳定,且明显优于多元回归分析结果.     

Knowledge Discovery from Medical Data and Development of an Expert System in Immunology

Artificial intelligence is one of the fastest-developing areas of science that covers a remarkably wide range of problems to be solved. It has found practical application in many areas of human activity, also in medicine. One of the directions of cooperation between computer science and medicine is to assist in diagnosing and proposing treatment methods with the use of IT tools. This study is the result of collaboration with the Children’s Memorial Health Institute in Warsaw, from where a database containing information about patients suffering from Bruton’s disease was made available. This is a rare disorder, difficult to detect in the first months of life. It is estimated that one in 70,000 to 90,000 children will develop Bruton’s disease. But even these few cases need detailed attention from doctors. Based on the data contained in the database, data mining was performed. During this process, knowledge was discovered that was presented in a way that is understandable to the user, in the form of decision trees. The best models obtained were used for the implementation of expert systems. Based on the data introduced by the user, the system conducts expertise and determines the severity of the course of the disease or the severity of the mutation. The CLIPS language was used for developing the expert system. Then, using this language, software was developed producing six expert systems. In the next step, experimental verification was performed, which confirmed the correctness of the developed systems.     

基于原子操纵技术的人工量子结构研究

扫描隧道显微镜原子操纵技术是指利用扫描探针在特定材料表面以晶格为步长搬运单个原子或分子的技术.它是纳米尺度量子物理与器件研究领域一种独特而有力的研究手段.利用这种手段,人们能够以原子或分子为单元构筑某些常规生长或微加工方法难以制备的人工量子结构,通过对格点原子、晶格尺寸、对称性、周期性的高度控制,实现对局域电子态、自旋序、以及能带拓扑特性等量子效应的设计与调控.原子操纵技术与超快测量及自动控制技术的结合,使得人们能够进一步研究原子级精准的量子器件,因而该技术成为探索未来器件新机理、新工艺的重要工具.本文首先简介原子操纵方法的发展过程和技术要点,然后分别介绍人工电子晶格、半导体表面人工量子点、磁性人工量子结构、人工结构中的信息存储与逻辑运算、单原子精度原型器件等方面的最新研究进展,以及单原子刻蚀和自动原子操纵等方面的技术进展,最后总结并展望原子操纵技术的应用前景和发展趋势.     

ML-Based Analysis of Particle Distributions in High-Intensity Laser Experiments: Role of Binning Strategy

When entering the phase of big data processing and statistical inferences in experimental physics, the efficient use of machine learning methods may require optimal data preprocessing methods and, in particular, optimal balance between details and noise. In experimental studies of strong-field quantum electrodynamics with intense lasers, this balance concerns data binning for the observed distributions of particles and photons. Here we analyze the aspect of binning with respect to different machine learning methods (Support Vector Machine (SVM), Gradient Boosting Trees (GBT), Fully-Connected Neural Network (FCNN), Convolutional Neural Network (CNN)) using numerical simulations that mimic expected properties of upcoming experiments. We see that binning can crucially affect the performance of SVM and GBT, and, to a less extent, FCNN and CNN. This can be interpreted as the latter methods being able to effectively learn the optimal binning, discarding unnecessary information. Nevertheless, given limited training sets, the results indicate that the efficiency can be increased by optimizing the binning scale along with other hyperparameters. We present specific measurements of accuracy that can be useful for planning of experiments in the specified research area.