Crystallography
1. Introduction
2. formulation of the proposed framework, 3. formulation of a multicomponent monodisperse spheres model, 4. numerical experiments, 5. discussion, 6. conclusions.
Format | | BIBTeX |
| | EndNote |
| | RefMan |
| | Refer |
| | Medline |
| | CIF |
| | SGML |
| | Plain Text |
| | Text |
|
research papers \(\def\hfill{\hskip 5em}\def\hfil{\hskip 3em}\def\eqno#1{\hfil {#1}}\)
| JOURNAL OF APPLIED CRYSTALLOGRAPHY |
Quantitative selection of sample structures in small-angle scattering using Bayesian methods
a Graduate School of Frontier Sciences, University of Tokyo, Kashiwa, Chiba 277-8561, Japan, b Japan Synchrotron Radiation Research Institute, Sayo, Hyogo 679-5198, Japan, c National Institute for Materials Science, Tsukuba, Ibaraki 305-0047, Japan, and d Facalty of Advanced Science and Technology, Kumamoto University, Kumamoto 860-8555, Japan * Correspondence e-mail: [email protected]
Small-angle scattering (SAS) is a key experimental technique for analyzing nanoscale structures in various materials. In SAS data analysis, selecting an appropriate mathematical model for the scattering intensity is critical, as it generates a hypothesis of the structure of the experimental sample. Traditional model selection methods either rely on qualitative approaches or are prone to overfitting. This paper introduces an analytical method that applies Bayesian model selection to SAS measurement data, enabling a quantitative evaluation of the validity of mathematical models. The performance of the method is assessed through numerical experiments using artificial data for multicomponent spherical materials, demonstrating that this proposed analysis approach yields highly accurate and interpretable results. The ability of the method to analyze a range of mixing ratios and particle size ratios for mixed components is also discussed, along with its precision in model evaluation by the degree of fitting. The proposed method effectively facilitates quantitative analysis of nanoscale sample structures in SAS, which has traditionally been challenging, and is expected to contribute significantly to advancements in a wide range of fields.
Keywords: small-angle X-ray scattering ; small-angle neutron scattering ; nanostructure analysis ; model selection ; Bayesian inference .
SAS measurement data are expressed in terms of scattering intensity that corresponds to a scattering vector, a physical quantity representing the scattering angle. Data analysis requires selection and parameter estimation of a mathematical model of the scattering intensity that contains information about the structure of the specimen. This selection process is critical as it involves assumptions about the structure of the specimen.
We conducted numerical experiments to assess the effectiveness of our proposed method. These experiments are based on synthetic data used to estimate the number of distinct components in a specimen, which was modeled as a mixture of monodisperse spheres of varying radii, scattering length densities and volume fractions. The results demonstrate the high accuracy, interpretability and stability of our method, even in the presence of measurement noise. To discuss the utility of the proposed method, we compare our approach with traditional model selection methods based on the reduced χ -squared error.
In this section, we present a detailed formulation of our algorithm for selecting mathematical models for SAS specimens using Bayesian model selection. The pseudocode for this algorithm is provided in Algorithm 1.
2.1. Bayesian model selection
The likelihood is thus expressed as
Let φ ( K ) be the prior distribution of the parameter K that characterizes the model, and φ ( Ξ | K ) be the prior distribution of the model parameters Ξ . Then, from Bayes' theorem, the posterior distribution of the parameters given the measurement data can be written as
2.2. Calculation of marginal likelihood
Sampling from the joint probability distribution at each inverse temperature gives
2.3. Estimation of model parameters
In this paper, we consider isotropic scattering and focus on the scattering vector's magnitude q , defined as
Monodisperse spheres are spherical particles of uniform radius. The scattering intensity I ( q , ξ ) of a specimen composed of sufficiently dilute monodisperse spheres of a single type for the scattering vector magnitude q is given by
To formulate the scattering intensity of a specimen composed of K types of monodisperse sphere, we assume a dilute system and denote the particle size of the k th component in the sample as R k and the scale as S k . The scattering intensity of a sample composed of K types of monodisperse sphere is then given by
| An illustration of a mixture of two types of spherical specimen. This shows scenarios with two components ( = 2), including mixtures of spherical particles of different sizes or volume fractions, and aggregates from a single particle type approximated as a large sphere. |
The numerical experiments reported in this section were conducted with a burn-in period of 10 5 and a sample size of 10 5 for the REMC. We set the number of replicas for REMC, the values of inverse temperature and the step size of the Metropolis method taking into consideration the state exchange rate and the acceptance rate.
4.1. Generation of synthetic data
(i) Set the number of data points to N = 400 and define the scattering vector magnitudes at N equally spaced points within the interval [0.1, 3] to obtain { q i } i =1 N =400 (nm −1 ).
In this section, we consider cases with pseudo-measurement times of T = 1 and T = 0.1. Generally, smaller values of T indicate greater effects from measurement noise.
4.2. Setting the prior distributions
In the Bayesian model selection framework, prior knowledge concerning the parameters Ξ and the model-characterizing parameter K is set as their prior distributions.
In this numerical experiment, the prior distributions for the parameters Ξ were set as Gamma distributions based on the pseudo-measurement time T used during data generation, while the prior for K was a discrete uniform distribution over the interval [1, 4].
| Plots of the prior distributions for various parameters. ( ) Prior distribution of , φ( ). ( ) Prior distribution of ) Prior distribution of , φ( ). ( ) Prior distribution of , φ( ). |
4.3. Results for two-component monodisperse spheres based on scale ratio
The ratio of the scale parameters S 1 and S 2 for spheres 1 and 2 during data generation, denoted r S , is defined as
Parameter values used for data generation with varying | | Sphere 1 | Sphere 2 | Radius (nm) | 2 | 10 | Scale | 250 | {250, 100, 20, 0.5, 0.1, 0.05} | Background (cm ) | 0.01 | Pseudo-measurement time | {1, 0.1} | | | Fitting to synthetic data generated at various values and residual plots. Panels and show cases for pseudo-measurement times of = 1 and = 0.1, respectively. In plots ( )–( ) and ( )–( ), the scale ratio is displayed in descending order for = 1 and = 0.1, respectively. Black circles represent the generated data and the black dotted lines indicate the true scattering intensity curves. For models = 1, = 2, = 3 and = 4, the fitting curves and residual plots are represented by blue dashed–dotted lines, red dashed lines, orange solid lines and green dotted lines, respectively. Fitting curves were plotted using 1000 parameter samples that were randomly selected from the posterior probability distributions for each model. The width of the distribution of these fitting curves reflects the confidence level at each point. | | Results of Bayesian model selection among models = 1–4 for varying values. Panel shows the posterior probability for each model using data generated with a pseudo-measurement time of = 1, and panel shows results for = 0.1. In cases ( )–( ) and ( )–( ), the scale ratio is displayed in descending order for = 1 and = 0.1, respectively. The height of each bar corresponds to the average values calculated for ten data sets generated with different random seeds, with maximum and minimum values shown as error bars. Areas highlighted in red indicate cases where, on average, the highest probability was found for the true model with = 2, while blue backgrounds indicate that models other than = 2 were associated with the highest probability on average. | The number of times each model was associated with the highest probability in numerical experiments for ten data sets generated with different random seeds at each value | | | | 1 | 2 | 3 | 4 | ( ) 1.0 | 0 | | 0 | 0 | ( ) 0.4 | 0 | | 0 | 0 | ( ) 0.08 | 0 | | 0 | 0 | ( ) 0.002 | 0 | | 0 | 0 | ( ) 0.0004 | 0 | | 0 | 0 | ( ) 0.0002 | | 2 | 0 | 0 | | | | | 1 | 2 | 3 | 4 | ( ) 1.0 | 0 | | 0 | 0 | ( ) 0.4 | 0 | | 0 | 0 | ( ) 0.08 | 0 | | 0 | 0 | ( ) 0.002 | 0 | | 0 | 0 | ( ) 0.0004 | | 1 | 0 | 0 | ( ) 0.0002 | | 0 | 0 | 0 | | 4.4. Results for two-component monodisperse spheres based on radius ratioDuring synthetic data generation, the ratio of the radii R 1 and R 2 of spheres 1 and 2, denoted r R , was defined as In this setup, we generated seven types of data by varying the value of r R for pseudo-measurement times of T = 1 and T = 0.1. Parameter values used for data generation when varying | | Sphere 1 | Sphere 2 | Radius (nm) | {9.9, 9.7, 9.5, 0.5, 0.5, 0.4, 0.3} | 10 | Scale | 250 | 100 | Background (cm ) | 0.01 | | Pseudo-measurement time | {1, 0.1} | | | | Fitting to synthetic data generated at various values and residual plots. Panels and show cases for pseudo-measurement times of = 1 and = 0.1, respectively. In plots ( )–( ) and ( )–( ), the radius ratio is displayed in descending order for = 1 and = 0.1, respectively. Black circles represent the generated data and the black dotted lines indicate the true scattering intensity curves. For models = 1, = 2, = 3 and = 4, the fitting curves and residual plots are represented by blue dashed–dotted lines, red dashed lines, orange solid lines and green dotted lines, respectively. Fitting curves were plotted using 1000 parameter samples that were randomly selected from the posterior probability distributions for each model. The width of the distribution of these fitting curves reflects the confidence level at each point. | | Results of Bayesian model selection among models = 1–4 for varying values. Panel shows the posterior probability of each model using data generated with a pseudo-measurement time of = 1, and panel shows results for = 0.1. In cases ( )–( ) and ( )–( ), the radius ratio is displayed in descending order for = 1 and = 0.1, respectively. The height of each bar corresponds to the average values calculated for ten data sets generated with different random seeds, with the maximum and minimum values shown as error bars. Areas highlighted in red indicate cases where the true model = 2 was most highly supported, while the blue backgrounds indicate that the likelihood of a model other than = 2 was the highest. | The number of times each model was most highly supported in numerical experiments for ten data sets generated by varying values | | | | 1 | 2 | 3 | 4 | ( ) 0.99 | | 1 | 0 | 0 | ( ) 0.97 | 0 | | 0 | 0 | ( ) 0.95 | 0 | | 0 | 0 | ( ) 0.5 | 0 | | 0 | 0 | ( ) 0.05 | 0 | | 0 | 0 | ( ) 0.04 | 1 | | 0 | 0 | ( ) 0.03 | | 0 | 0 | 0 | | | | | 1 | 2 | 3 | 4 | ( ) 0.99 | | 0 | 0 | 0 | ( ) 0.97 | 2 | | 0 | 0 | ( ) 0.95 | 0 | | 0 | 0 | ( ) 0.5 | 0 | | 0 | 0 | ( ) 0.05 | 1 | | 0 | 0 | ( ) 0.04 | | 3 | 0 | 0 | ( ) 0.03 | | 0 | 0 | 0 | | 5.1. Limitations of the proposed method5.2. model selection based on χ -squared error. In SAS data analysis, selecting an appropriate mathematical model for the analysis is a crucial but challenging process. In this subsection, we compare the conventional model selection method based on the χ -squared error with the results of model selection using our proposed method. | The fitting results and residual plots for the data shown in Fig. 3 ( ) were derived using parameters that minimize the χ-squared error from the posterior probability distributions for models ranging from = 1 to = 4. For each of these models, the fitting curves and their corresponding residual plots are represented by blue dashed–dotted lines, red dashed lines, orange solid lines and green dotted lines, respectively. The legend indicates the reduced χ-squared values for each model ( = 1 to = 4). | Model selection results based on reduced χ-squared values | -squared value to 1 for ten data sets generated with different random seeds for each setting = 1. Labels ( ) to ( ) refer to the settings in Figs. 3–4 and Table 2. The cases with the highest level of support for each data set are shown in bold. | | | | 1 | 2 | 3 | 4 | ( ) 1.0 | 0 | 2 | | 0\sim | ( ) 0.4 | 0 | 0 | | 1 | ( ) 0.08 | 0 | 0 | | 1 | ( ) 0.002 | 0 | 0 | | 0 | ( ) 0.0004 | 0 | 4 | | 1 | ( ) 0.0002 | 0 | 2 | | 0 | | In this paper, we have introduced a Bayesian model selection framework for SAS data analysis that quantitatively evaluates model validity through posterior probabilities. We have conducted numerical experiments using synthetic data for a two-component system of monodisperse spheres to assess the performance of the proposed method. We have identified the analytical limits of the proposed method, under the settings of this study, with respect to the scale and radius ratios of two-component spherical particles, and compared the performance of traditional model selection methods based on the reduced χ -squared. The numerical experiments and subsequent discussion reveal the range of parameters that can be analyzed using the proposed method. Within that range, our method provides stable and highly accurate model selection, even for data with significant noise or in situations in which qualitative model determination is challenging. In comparison with the traditional method of selecting models based on fitting curves and data residuals, it was found that the proposed method offers greater accuracy and stability. SAS is used to study specimens with a variety of structures other than spheres, including cylinders, core–shell structures, lamellae and more. The proposed method should be applied to other sample models to determine the feasibility of expanding the analysis beyond the case examined here to broader experimental settings. Future work could benefit from using the proposed method to conduct real data analysis, which is expected to yield new insights through our more efficient analysis approach. Funding informationThis work was supported by JST CREST (grant Nos. PMJCR1761 and JPMJCR1861) from the Japan Science and Technology Agency (JST) and by a JSPS KAKENHI Grant-in-Aid for Scientific Research (A) (grant No. 23H00486). This is an open-access article distributed under the terms of the Creative Commons Attribution (CC-BY) Licence , which permits unrestricted use, distribution, and reproduction in any medium, provided the original authors and source are cited. Follow J. Appl. Cryst. | Research on abnormal diagnosis model of electric power measurement based on small sample learning- Zhuang, Gewei
- Zhang, Jingyue
- Zhang, Honghong
For a long time, abnormal metering of electricity meters has caused huge economic losses to power grid companies. Abnormal diagnosis of power metering is an important means to ensure the normal operation of electricity meters and power automation operation and maintenance systems and is a hot topic of research for power workers. This article proposes a known measurement anomaly diagnosis model based on small sample learning to address the problem of insufficient labeled samples in power measurement anomaly diagnosis. The embedded network maps samples from the original sample space to the embedded space adjusts the embedded network structure, and improves the loss function. The experimental results show that the improved classification network has a higher recognition accuracy for known anomalies than the original network and other small sample learning models. |
| | | | |
COMMENTS
Writing the Experimental Report: Methods, Results, and Discussion. Tables, Appendices, Footnotes and Endnotes. References and Sources for More Information. APA Sample Paper: Experimental Psychology. Style Guide Overview MLA Guide APA Guide Chicago Guide OWL Exercises. Purdue OWL. Subject-Specific Writing.
Step 1: Define your variables. You should begin with a specific research question. We will work with two research question examples, one from health sciences and one from ecology: Example question 1: Phone use and sleep. You want to know how phone use before bedtime affects sleep patterns.
Examples of Experimental Research. 1. Pavlov's Dog: Classical Conditioning. Pavlovs Dogs. Dr. Ivan Pavlov was a physiologist studying animal digestive systems in the 1890s. In one study, he presented food to a dog and then collected its salivatory juices via a tube attached to the inside of the animal's mouth.
Sample One-Experiment Paper (continued) emotional detection than young adults, or older adults could show a greater facilitation than. young adults only for the detection of positive information. The results lent some support to the. first two alternatives, but no evidence was found to support the third alternative.
Experimental research serves as a fundamental scientific method aimed at unraveling. cause-and-effect relationships between variables across various disciplines. This. paper delineates the key ...
Methods. Using an experimental design we photographed the faces of 23 adults (mean age 23, range 18-31 years, 11 women) between 14.00 and 15.00 under two conditions in a balanced design: after a normal night's sleep (at least eight hours of sleep between 23.00-07.00 and seven hours of wakefulness) and after sleep deprivation (sleep 02.00-07.00 and 31 hours of wakefulness).
Experimental reports (also known as "lab reports") are reports of empirical research conducted by their authors. You should think of an experimental report as a "story" of your research in which you lead your readers through your experiment. As you are telling this story, you are crafting an argument about both the validity and reliability of ...
1) True Experimental Design. In the world of experiments, the True Experimental Design is like the superstar quarterback everyone talks about. Born out of the early 20th-century work of statisticians like Ronald A. Fisher, this design is all about control, precision, and reliability.
Step 1: Define your variables. You should begin with a specific research question. We will work with two research question examples, one from health sciences and one from ecology: Example question 1: Phone use and sleep. You want to know how phone use before bedtime affects sleep patterns.
February 2011. by Jeff Galak and Tom Meyvis. The Nature of Gestures' Beneficial Role in Spatial Problem Solving (PDF, 181KB) February 2011. by Mingyuan Chu and Sotaro Kita. Date created: 2009. Sample articles from APA's Journal of Experimental Psychology: General.
Three types of experimental designs are commonly used: 1. Independent Measures. Independent measures design, also known as between-groups, is an experimental design where different participants are used in each condition of the independent variable. This means that each condition of the experiment includes a different group of participants.
You can also create a mixed methods research design that has elements of both. Descriptive research vs experimental research. Descriptive research gathers data without controlling any variables, while experimental research manipulates and controls variables to determine cause and effect.
A proper experimental design serves as a road map to the study methods, helping readers to understand more clearly how the data were obtained and, therefore, assisting them in properly analyzing the results. Keywords: scientific writing, scholarly communication. Study, experimental, or research design is the backbone of good research.
Experimental design is the process of carrying out research in an objective and controlled fashion. so that precision is maximized and specific conclusions can be drawn regarding a hypothesis ...
There are 3 types of experimental research designs. These are pre-experimental research design, true experimental research design, and quasi experimental research design. 1. The assignment of the control group in quasi experimental research is non-random, unlike true experimental design, which is randomly assigned. 2.
This article aims to present general guidelines to one of the many roles of a neurosurgeon: Writing an experimental research paper. Every research report must use the "IMRAD formula: introduction, methods, results and discussion". After the IMRAD is finished, abstract should be written and the title should be "created".
A research paper is intended to inform others about advancement in a particular field of study. The researcher who wrote the paper identified a gap in the research in a field of study and used their research to help fill this gap. The researcher uses their paper to inform others about the knowledge that the results of their study contribute ...
The results have been fed into SPSS (12.0) and analyzed using independent sample T-test analysis. Table 2 shows that in Test 1, Group 1 and Group 2 are quite similar in the means (Group 1 is 69.33, while Group 2 is 70.92), this means both groups have nearly the same English proficiency, and though experimental group is a little
experimental group that was class XI.2 as an experimental group (46 students) and class XI.5 as a control group (46 students). It means tha t totally 92 students were the sample of the resea rch.
The three main types of experimental research design are: 1. Pre-experimental research. A pre-experimental research study is an observational approach to performing an experiment. It's the most basic style of experimental research. Free experimental research can occur in one of these design structures: One-shot case study research design: In ...
Here are reminders on how you could improve your research writing skills. Who knows, one day, you will join the ranks of world changers with your experimental research report. 1. Identify the Problem. To solve a problem, you need to define what it is first. You can begin with identifying the field of research you wish to investigate, then find ...
A real-world example of experimental research is Pavlov's Dog experiment. In this experiment, Ivan Pavlov demonstrated classical conditioning by ringing a bell each time he fed his dogs. After repeating this process multiple times, the dogs began to salivate just by hearing the bell, even when no food was presented. ...
The analysis and interpretation of data is carried out in two phases. The. first part, which is based on the results of the questionnaire, deals with a quantitative. analysis of data. The second, which is based on the results of the interview and focus group. discussions, is a qualitative interpretation.
Compute-efficient training of large language models (LLMs) has become an important research problem. In this work, we consider data pruning as a method of data-efficient training of LLMs, where we take a data compression view on data pruning. We argue that the amount of information of a sample, or the achievable compression on its description length, represents its sample importance. The key ...
The experimental method formally surfaced in educational psy-. chology around the turn of the century, with the classic studies. by Thorndike and Woodworth on transf er (Cronbach, 1957). The ...
This paper introduces an analytical method that applies Bayesian model selection to SAS measurement data, enabling a quantitative evaluation of the validity of mathematical models. The performance of the method is assessed through numerical experiments using artificial data for multicomponent spherical materials, demonstrating that this ...
For a long time, abnormal metering of electricity meters has caused huge economic losses to power grid companies. Abnormal diagnosis of power metering is an important means to ensure the normal operation of electricity meters and power automation operation and maintenance systems and is a hot topic of research for power workers. This article proposes a known measurement anomaly diagnosis model ...