Search

Data science is a science that is currently very popular and can be applied in various industries. Of course, the need for proficient people in this field is growing in demand by many businesses, including large corporations. Data science is a science that combines expertise in a particular field of science with programming, mathematics, and statistics skills. The goal is to extract knowledge or information from data. The problem that usually occurs in the data science process is at the data modeling stage, namely the accuracy of the mathematical model that will be used to obtain the estimated parameter values. To overcome this problem, this research was conducted to test the correctness of analytic derivatives in mathematical models and conduct data modeling using analytic derivatives that have been tested to obtain the estimated value of each parameter. The validation process is carried out using an analytic and numeric gradient comparison. If the comparison of the two values is close to zero, then the analytical derivatives used are correct. The parameter estimation method used is the Maximum Likelihood Estimation. In this research, it was concluded that the value of the analytic gradient is the same or close to the numeric gradient value for each parameter so that the analytic derivatives are correct and can be used for the next stage. With the results of this research, the expected implication is to get more accurate data modeling so that the conclusions obtained are more precise. [ABSTRACT FROM AUTHOR]

Embedding technology in teaching and learning (T&L) becomes necessary, especially during open distance learning (ODL). Mobius is a math-based online platform that provides interactive T&L applications to meet the ODL requirements. This platform is built on Maple software with intelligent grading and response, where students can attempt several random questions within the same subtopic and get an immediate response. However, students' response to this platform is the primary concern. Thus, this study analyzed the students' feedback on using Mobius for Calculus I. The content designs, implementation, and students' experience using this platform are discussed. The respondents in this study were 20 non-mathematical students in the second semester of Bachelor of Information Technology, Faculty of Computer and Mathematical Sciences, UiTM Shah Alam. Students were given access to Mobius and required to complete all the exercises on selected topics in Calculus I - techniques of differentiation, chain rule, techniques of integration and application of integration. Finally, their feedback is recorded and analyzed. The result shows that most students showed positive feedback on using Mobius to improve their Calculus I understanding and recommended using Mobius in T&L sessions. [ABSTRACT FROM AUTHOR]

The paper presents the analysis results of scientometric data of modern Russian research in Computer science to form and evaluate thematic trends, as well as determine their position relative to the trends of world scientific publications. The study is based on the methods of scientometric analysis of bibliographic data to reveal the thematic distribution of scientific publications, including in retrospect. The bibliographic database of the scientific electronic library eLIBRARY.Ru, the Scopus database, as well as the annual reports "Research fronts" published by Clarivate Analytics were used as data sources. In the course of the study, an examination of scientific publications was carried out, and internal research fronts were identified. A conclusion is drawn about the catching-up position of Russian research in the field of Computer sciences, and an assumption is made about the thematic areas that will be relevant in Russia in the coming years. [ABSTRACT FROM AUTHOR]

This research was conducted to obtain Faculty Quality assurance at the Faculty of Computer Science, Santo Thomas Catholic University. The scope includes the vision and mission, curriculum, journals and research on lecturers. The Faculty's Vision and Mission and the Study Program's Vision and Mission aim to have future directions and goals to be better and achieve for the future of the Faculty of Computer Science. RAD is a combination of various structured techniques with prototyping techniques and joint application development techniques to accelerate system or application development. From the definitions of this RAD concept, it can be seen that application development using the RAD method can be done in a relatively faster time. With the quality assurance application system at the Faculty of Computer Science, Santo Thomas Catholic University, it can be made efficiently. Programs that are made can be implemented for assessment and arranged in such a way as to be sequential values that make faculty quality guidelines. [ABSTRACT FROM AUTHOR]

In this paper an advanced stochastic approach based a lattice rule with improved generating vector has been obtained and studied. The Unified Danish Eulerian Model (UNI-DEM) is a key element in our investigation since it is one of the most advanced large-scale mathematical models that describes adequately all physical and chemical processes. The calculations are done in a large spatial domain, which covers completely the European region and the Mediterranean, for certain time period. What is more, it studies the transportation of a number of pollutants and all these make it a perfect case study. The paper compares the described lattice rule and demonstrates its superiority over the plain Monte Carlo approach and Fibonacci based lattice rule. The proposed approach is robust and computationally efficient in estimating multidimensional integrals and thus sensitivity indices, which is vital in scientific computing. [ABSTRACT FROM AUTHOR]

This article, as an example of the integration of disciplines, is focused on the issues of combining knowledge, skills and practical experience at all levels of training of software engineering specialists and synthesizing knowledge directed to specific field goals are discussed. At the same time, modern information and communication technologies are considered as a means of integration that connects mathematics and computer sciences with each other. [ABSTRACT FROM AUTHOR]

Cloud users and service providers are the leading players in the cloud computing environment. This environment comprises data centers, hosts, agents and virtual machines. The cloud user's application of varied loads is leased on the provider's resources. Scientific applications are large-scale complex workflow problems that demand more computing power. The cloud fulfills the workflow requirements of huge availability and increased computational power. One of the most crucial issues of cloud computing is scheduling tasks for the system's effective functioning. This paper reviews several existing task-scheduling techniques based on diverse metrics. This work will help the investigators to gain a better understanding of task scheduling techniques. In order to boost an algorithm's performance, a few strategies are offered. [ABSTRACT FROM AUTHOR]

The present situation is investigated and the relevance of the stage of prototyping of user interfaces as intermediary of human-machine interaction in the process of software development is substantiated. The necessity of improving the methodology of teaching graphic disciplines and the relevance of studying the technologies of «UX/UI Design» in the process of professional training of future bachelor's in computer science as a perspective direction in view of the demands of the labor market and as a factor that raises the level of motivation of students for professional activity in general and on the formation of graphic competence. The methodical provision of the course is described, the structure and content of the theoretical and practical part of the course are presented, which is focused on the study of the technology of «UX/UI Design». The mathematical modeling of the discipline learning process «UX/UI Design» is described. The use of mathematical modeling to develop a model for training the discipline «UX/UI Design» provided the basis for the creation of a system of scientific-methodical and computer support of pedagogical activities and outlined the perspectives of designing a pedagogical technology for the formation of graphic competence of future Bachelor of Computer Sciences in the course of studying the discipline «UX/UI Design». [ABSTRACT FROM AUTHOR]

The article examines probabilistic logic as applied to computer science and its problems. Probabilistic logic is treated as a generalization of formal logic. In many cases, the application of probabilistic logic occurs under conditions of uncertainty. Probabilistic logic uses a model approach. Computer science also uses a model approach. Probabilistic logic describes uncertainty using quantitative values of probability. Computer science describes uncertainty using the concept of information entropy. The article introduces the concept of informational logical uncertainty. The article finds a connection between probabilistic logic and entropy. This connection makes it possible to build a probabilistically logical model. This connection makes it possible to create probabilistically logical modeling, which is in many ways an analogue of information modeling. The purpose of probabilistically logical modeling is to overcome information and logical uncertainty. [ABSTRACT FROM AUTHOR]

Big Data has become the new slang in the world of information collection and analysis. The researches we conduct and the data we collect continue to grow, due to rapidly expansion of technology. Disciplines such as Computer Science, Engineering, and Statistics play a key role in the analysis of big data, each with its specificity but all equally important, an opinion that is not shared by all, being the Statistic considered the weakest link. This work attempts to show that Statistics have a distinct and essential role in this new world of Big Data, showing that Statistics and Big Data denote a crucial union. We will start with a brief introduction to Big Data and the several existing definitions. [ABSTRACT FROM AUTHOR]

Sentiment analysis is a widely researched area in computer science that have been attracting researchers since the past few decades. Shift in polarity is a problem that exists in sentiment classification. Another problem is the difficulty in handling inconsistent sentiment polarity between a phrase and its constituent words. For solving both, we perform dual sentiment analysis using an original dataset and its reversed version. We use a sentiment classification approach for classifying the sentiments of the reviews in both the original and reversed sets. This approach is based on segmenting the sentences and then using the segmentation results to find the polarity of the sentences. Then rules are mined to correctly classify the reviews. When this approach is applied on a product review dataset taken from amazon.com, it shows good precision, recall and f1-score values. [ABSTRACT FROM AUTHOR]

Complexity theory is concerned with estimating the (generally worst-case) amount of computational resources (time, memory, etc.) required to solve different problems, while scientific computing and numerical analysis are concerned with computing given functions on (generally error-contaminated) input data and estimating bounds on the precision with which different algorithms can perform these tasks. The concept of computational intractability describes tasks which appear to require inordinate amount of computing effort in the worst case. I here argue that this has a counterpart in numerical analysis, namely, ill-conditioning of data problems, an idea introduced by Alan Turing in 1948. By converting the classical heat equation to a numerical backwards problem, I show that the resulting system Mb = y is exponentially badly conditioned in the sense that the matrix M has an exponentially exploding condition number due to eigenvalues collapsing near an accumulator at 0. This example suggests a link between ideas (and problems) from complexity theory and numerical analysis and hints at a deep duality between computational intractability and the irreversibility of the arrow of time. [ABSTRACT FROM AUTHOR]

The aim of this research is to improve science process skills and students cognitive learning outcomes, also to find out the correlations between these variables. The research method is Classroom Action Research (CAR) based on Lesson Study (LS) in applying guided inquiry learning models. This research uses a correlational descriptive approach—the data of science process skills and students cognitive learning outcomes collected using written test instruments. Learning implementation data is obtained based on observation. CAR-LS is conducted in two cycles. Classroom Action Research consists of four stages are planning, implementing, observing, and reflecting. In one cycle of each meeting is carried out by applying stages of Lesson Study that are plan, do, and see. Cycle I was carried out on Human Digestive Systems in eight meetings. Cycle II was carried out on Human Respiratory Systems in six meetings. The subject of this research are students of Grade 11th in Science Major SMAN 4 Malang at 2nd Academic Year 2019/2020 of about 32 students consist of 20 female students and 14 male students. The results showed that N-gain of science process skills increased from 0.6418 in cycle I to 0.8941 in cycle II. N-gain cognitive learning outcomes increased from 0.6491 in cycle I to 0.8199 in cycle II. The value of correlation coefficient (r) between science process skills and cognitive learning outcomes is 0.529. The conclusion from the results is the implementation of guided inquiry learning models can improve science process skills and students cognitive learning outcomes. In addition, there is a correlation between science process skills and cognitive learning outcomes with a moderate level of correlation. [ABSTRACT FROM AUTHOR]

This study aims to determine: (1) the profile of science process skills and critical thinking skills of prospective chemistry teachers in terms of self-regulated learning, (2) a description of science process skills and critical thinking skills of the first and second-year of prospective chemistry teachers, and (3) the relationship between science process skills and critical thinking skills of prospective chemistry teachers. The design of this research is survey research. The research subjects were the first and second-year of prospective chemistry teachers at a State University in Indonesia. The instruments used were: (1) science process skills test (reliability, r = 0.692), (2) critical thinking skills test (r = 0.678), and (3) self-regulated learning questionnaire (r = 0.705). Data on science process skills and critical thinking skills of prospective chemistry teachers were analyzed descriptively and inferential statistics. The results showed that: (1) most of the students had science process skills in the high category, while their critical thinking skills were in the moderate category, (2) most students with high self-regulated learning had high science process skills, but students with high critical thinking had low self-regulated learning, (3) there is no difference in science process skills and critical thinking skills between the first and second-year students, (4) there is a relationship between students' science process skills and critical thinking skills (correlation r = 0.522). [ABSTRACT FROM AUTHOR]

This paper presents the up-to-date results concerning the integration of scientific classification and computation deploying long-term knowledge resources. The knowledge resources are the central component in universal multi-disciplinary and multi-lingual knowledge management. While the research focus is on universal knowledge creation, documentation, and preservation, the architecture allows efficient dynamical use and processing as well as computational utilisation, e.g., in advanced scientific computing with Integrated Information and Computing System (IICS) components and High End Computing (HEC) resources. The knowledge resources' goal is to address a major gap for long-term knowledge and to provide structure, references, and classification for any type of data, knowledge, and context. Based on the implementation, the case study depicts concepts of references to Universal Decimal Classification (UDC) views, e.g., for workflows. The key results are benefits for knowledge management and documentation, the sustainable integration of long-term documentation, data, and workflow vitality for improved knowledge gathering, computational purposes, and evaluation based on big data. [ABSTRACT FROM AUTHOR]

In this paper, the Restricted additive Schwarz (RAS) method is applied to solve Helmholtz equation. To accelerate the RAS iterations, we propose to apply the vector "ε-algorithm. Some convergence analysis of the proposed method is presented, and applied succeffully to Helmholtz problem. The obtained results show the efficiency of the proposed approach. Moreover, the algorithm yields much faster convergence than the classical Schwarz iterations. [ABSTRACT FROM AUTHOR]

We summarize the issues and concerns voiced by the ACAT conference attendees regarding their experiences using the C++ programming language and attendant methodologies. © 2001 American Institute of Physics. [ABSTRACT FROM AUTHOR]

This paper presents the recent results of the integration of computational and conceptual solutions for the complex case of environmental information management. The solution for the major goal of creating and developing long-term multi-disciplinary knowledge resources and conceptual and computational support was achieved by implementing and integrating key components. The key components are long-term knowledge resources providing required structures for universal knowledge creation, documentation, and preservation, universal multi-disciplinary and multi-lingual conceptual knowledge and classification, especially, references to Universal Decimal Classification (UDC), sustainable workflows for environmental information management, and computational support for dynamical use, processing, and advanced scientific computing with Integrated Information and Computing System (IICS) components and High End Computing (HEC) resources. [ABSTRACT FROM AUTHOR]

Information about the topics discussed at a symposium entitled "Recent Developments in Hilbert Space Tools and Methodology for Scientific Computing" is presented. Topics include the high dimensional model representation (HDMR) for multivariate functions, fluctuation terms and related techniques in integration and matrix representation, and fine techniques in quadratures via Hilbert space concepts. In addition, the relation of fluctuation concept with quantum dynamical problems was tackled.

This paper presents the results based on the research from the integration of advanced methods formerly used in different disciplines and its practical implementation. The integration focusses on content description on the one hand and on computational methods on the other hand. The result of the integration enables advanced knowledge mining workflows, spanning the knowledge of many conceptual knowledge implementations. The integration includes classifications implementing conceptual knowledge, content and classification based verbal descriptions, and methods used with knowledge mining computation. Conceptual knowledge is a rich source of high quality if used appropriately. The value of conceptual knowledge references even increases when considering verbal descriptions, which are available for most implementations. The goal of this research is to integrate conceptual knowledge and associated verbal description in the knowledge mining process. In the context of complex knowledge resources, referencing to multi-disciplinary long-term knowledge, the exploitation can lead to much improved mining results and new insights. The integration goes along with challenges for the creation of resources and design of High End Computation (HEC) workflows. [ABSTRACT FROM AUTHOR]

This paper presents research based on the new Content Factor (CONTFACT) applied to long-term knowledge resources, which are continuously in development. The CONTFACT methodology was especially created for data description and analysis complementary to previously existing methods. The Content Factor method can be applied to arbitrary data and content and it can be adopted for many purposes like data analysis and knowledge discovery. The goal of this research is to use the new methodology in order to create additional data-centric description to multi-disciplinary long-term knowledge resources. The new methodology supports flexible ways for data description and analysis and can be used with huge structured and even unstructured data resources, e.g., with Integrated Information and Computing System (IICS) components and High End Computing (HEC) resources. [ABSTRACT FROM AUTHOR]

Scientific computing has become a tool as vital as experimentation and theory for dealing with scientific challenges of the twenty-first century. Large scale simulations and modelling serve as heuristic tools in a broad problem-solving process. High-performance computing facilities make possible the first step in this process — a view of new and previously inaccessible domains in science and the building up of intuition regarding the new phenomenology. The final goal of this process is to translate this newly found intuition into better algorithms and new analytical results. In this presentation we give an outline of the research themes pursued at the Scientific Computing Laboratory of the Institute of Physics in Belgrade regarding large-scale simulations of complex classical and quantum physical systems, and present recent results obtained in the large-scale simulations of granular materials and path integrals. © 2007 American Institute of Physics [ABSTRACT FROM AUTHOR]

Coronavirus (Covid-19) is an infectious disease that attacks the respiratory tract caused by the newly discovered corona virus. In dealing with the spread of covid-19, WHO has declared quarantine and isolation and carried out vaccinations. The implementation of quarantine and isolation during the vaccination's period is expected to suppress the spread of covid-19. The purpose of the study was to determine the effectiveness of quarantine and isolation during the vaccination period in suppressing the spread of covid-19. In this study, an analysis of the effectiveness of quarantine and isolation against the spread of covid-19 disease was carried out during the vaccination's period using the SEQIJR model. The SEQIJR model is in accordance with the problems that occur, which involve 6 classes are susceptible (S), exposed (E), quarantined (R), infectious (I), isolated (J), and recovered (R). Through this model, there are two equilibrium points, namely the disease-free equilibrium point and the endemic equilibrium point. The disease-free equilibrium point (ε0) is locally asymptotically stable with R0 < 1 and the endemic equilibrium point ( ε 0 * ) is unstable. This has the interpretation that the spread of covid-19 will disappear from the population over time. The simulation shows that tightening interventions in the form of quarantine and isolation can accelerate the decline in the rate of spread of covid-19. Meanwhile, isolation interventions are better than quarantine in suppressing the spread of covid-19. [ABSTRACT FROM AUTHOR]

A Monte Carlo method for modeling of a viral epidemic is presented in the paper. Two modifications of the model are discussed. At the first modification random numbers determine the contacts of one infective and the probability for transmission of the virus. At the second modification randomly are determined the number and the days of the secondary infections. The two modifications give equivalent results but the second one is numerically much more effective. The model is validated by comparison with other methods. Some test results of the model are also presented in the paper. [ABSTRACT FROM AUTHOR]

In this work, the annual cycle and interannual variability of the content of greenhouse gases were studied in the Arctic regions of Russia. Gases concentrations (methane, water vapor) and main meteorological parameters (temperature, atmospheric pressure) were measured on the Arctic Island Belyy, Russia in July-August, 2015-2017. The daily mean concentrations of greenhouse gases were not differed within the mean plus or minus standard deviation year to year. The autoregressive models and the spectral Fourier analysis were built for determining the pattern of the dynamic changes in the time series Spearman's rank correlation coefficients between the greenhouse gases concentrations and temperature. The analysis of the dependence of Spearman's rank correlation coefficients on the time delay of the temperature reaction to a change in the content of greenhouse gases in the air revealed four invariants: one, three, seven, and ten days. There were explanations for the daily, weekly and ten-day rhythms, but the three-day one requires further explanation. [ABSTRACT FROM AUTHOR]

This paper presents the results from the development and implementation of algorithms for diffraction amplitudes for wave phenomena to be dynamically used with Integrated Information and Computing System (IICS) components and High End Computing (HEC) resources. The key issue is the principle of computing diffraction amplitudes, e.g., with seismic data processing and how an algorithm can be used with interactive systems on supercomputing resources, supporting applications for advanced scientific computing as well as improved analysis and interpretation support with dynamical Geoscientific and Archaeological Information Systems (GIS, AIS). Further, the components created are considered objects of sustainable knowledge resources for long-term persistent big data vitality of documentation, processing, analysis, and evaluation. [ABSTRACT FROM AUTHOR]

The arrival of hardware accelerators has created a new gold rush to be the first to deliver their promise of high performance for numerical applications. Despite the recent advances in programmability, it is still hard to develop tuned programs that extract all the potential performance promised by the manufacturers. In this paper we remind the community that while this development effort is a noble endeavor, there is a lot of low hanging fruit that can be harvested easily. Picking this low hanging fruit benefits the scientific computing community immediately and prototypes the approach that further optimizations may follow. We demonstrate this by focusing on a widely used set of operations, the level-3 BLAS, targeting the NVIDIA GPUs. [ABSTRACT FROM AUTHOR]

This paper presents the results from the development and implementation of Superstack algorithms to be dynamically used with integrated systems and supercomputing resources. Processing of geophysical data, thus named geoprocessing, is an essential part of the analysis of geoscientific data. The theory of Superstack algorithms and the practical application on modern computing architectures was inspired by developments introduced with processing of seismic data on mainframes and within the last years leading to high end scientific computing applications. There are several stacking algorithms known but with low signal to noise ratio in seismic data the use of iterative algorithms like the Superstack can support analysis and interpretation. The new Superstack algorithms are in use with wave theory and optical phenomena on highly performant computing resources for huge data sets as well as for sophisticated application scenarios in geosciences and archaeology. [ABSTRACT FROM AUTHOR]

The open-source software R has become one of the most widely used systems for statistical data analysis and for making graphs, but it is also well suited for other disciplines in scientific computing. One of the fields where considerable progress has been made is the solution of differential equations. Here we first give an overview of the types of differential equations that R can solve, and then demonstrate how to use R for solving a 2-Dimensional partial differential equation. [ABSTRACT FROM AUTHOR]

In this work we present results of numerical simulation of three-dimensional temperature fields in soils for various applied problems: the railway line in the conditions of permafrost for different geometries, the horizontal tunnel underground storage and greenhouses of various designs in the Far North. Mathematical model of the process is described by a nonstationary heat equation with phase transitions of pore water. The numerical realization of the problem is based on the finite element method using a library of scientific computing FEniCS. For numerical calculations we use high-performance computing systems. [ABSTRACT FROM AUTHOR]

This article presents the results of 2D numerical simulations to estimate the change of local seismic response due to an excavation that occurred during the L’Aquila city’s earthquake (Italy) in April 2009. The modeling is an attempt to understand the very complex phenomenon produced by multiple reflections due to particular topographic and stratigraphic geometries, verifying whether the anthropic actions are able to modify or not the surface amplifications. In particular, the numerical simulations were performed considering national and international earthquakes databases, compatible to the spectral form expected at the site. The investigations were conducted with the VisualQ4M software that is a Pre & Post Processor capable to use up in the background the Quad4M calculation code for the study of 2D seismic amplifications. In the numerical context, the model was reconstructed according to the natural slopes and topographical conformations of the site and of the stratigraphic discontinuities, using a structured mesh with dimension suitable for the minimum speed of propagation of seismic waves (Vs) in order to avoid filtering or attenuations of wave propagation. [ABSTRACT FROM AUTHOR]

This work deals with the issues specific to the development of a massively parallel topology optimization solver. This is developed as a module in the open source software ExaScale PaRallel FETI SOlver (ESPRESO, http://espreso.it4i.cz/). In a topology optimization problem, the heterogeneity of the system increases, which adversely affects the performance of the Finite Element Tearing and Interconnection (FETI) method. Robust FETI variants which can handle such heterogeneities are discussed here. Their performance is tested and the results are presented. A suitable FETI method is identified for implementation in topology optimization solver. [ABSTRACT FROM AUTHOR]

Robust methods, such as Tikhonov regularization and Bounded data uncertainty, have been used extensively in relatively small problems involving dense matrices for many decades, but have not been used in large-scale iterative methods for image reconstruction in particle imaging until recently. In this case, robust methods may allow more accurate reconstruction of images in the presence of errors of both the energy measurement of the protons and ions but also in the estimated path taken by the proton or ion through the object. Robust systems may also be used when entire blocks of data are missing, or in low-dose reconstructions using a very small number of particles without substantial loss of image quality. In this contribution, we demonstrate that robust methods show great promise in proton/ion (particle) computed tomography (pCT), and, for the first time, that they can be proven to converge. Thus, the convergence of robust methods as well as benefits for reconstruction in uncertain systems is shown to constitute the main advantage for pCT reconstruction. [ABSTRACT FROM AUTHOR]

Computational simulations of multiscale deformable porous media are routinely encountered as a part of research and development activities in a number of engineering, environmental and biomedical fields. The efficiency of multilevel iterative solution of such problems is a challenging topic on numerical methods for large-scale scientific computing, this is because predicting the mechanical behavior of such systems with hierarchical structures with multiple scales is very computationally demanding. Our main interest application concerns medium that has complex hierarchical morphology in the sense that features ranges from nanometer to millimeter scales. The goal of this work is to propose a computationally efficient numerical tool that can be used to perform everyday predictive simulations as an integral part of osteoporosis treatment, for example. To achieve that, highly heterogeneous media are considered that resembles trabecular bone tissues. The related fine-scale linear elasticity problem is of high contrast and high frequency. The finite element method (FEM) is applied for discretization of the related linear elasticity problem, using separable displacement decomposition. The new feature in this work is that at coarser levels, a block diagonal preconditioner is applied that incorporates an analytical effective tensor into the simulation, avoiding costly numerical solutions of local problems that are inherent in methods for multiscale problems. The robustness of the new proposed algorithm is measured by comparing the number of V-cycles necessary to resolve the considered multiscale problems with other well known techniques. [ABSTRACT FROM AUTHOR]

It is well known that eddy current measurements of cracks in titanium are corrupted by the presence of grain noise, which is generated due to local changes in the conductivity and crystal orientation of the material. This grain noise is spatially correlated and may complicate the inverse problem, potentially leading to inaccurate estimates of the crack geometry if it is not modeled properly. As has been demonstrated in [1], the grain noise can be modeled as a Gaussian process (GP), using spectral mixture kernels to represent the covariance function. However, in [1], the GP model used in the inversion was fit to model discrepancy data, which was found by subtracting computed model signals from specimens with known crack sizes. It is shown here that it is sufficient to fit the GP model to grain noise data only. Furthermore, unlike in [1], where a representative covariance function was applied in all inversions, the covariance function is re-estimated on a case-by-case basis. As in [1], the inverse problem was solved within the Bayesian framework. In addition to computing the MAP estimates of the unknowns assuming a uniform prior, a novel prior is proposed for the length-to-depth ratio of the crack that eliminates unlikely solutions. [ABSTRACT FROM AUTHOR]