Core Machine Research Articles

Machine Learning Library in C++

This project aims to develop a comprehensive and versatile machine learning library in C++ tailored to address the diverse needs of developers and researchers in the field. The library encompasses a robust set of core machine learning algorithms, encompassing supervised, unsupervised, and reinforcement learning techniques. Additionally, it incorporates essential data preprocessing tools to streamline data manipulation and feature engineering tasks, along with model evaluation capabilities crucial for assessing algorithm performance. The library's primary focus is on providing a rich suite of machine learning algorithms, This empowers users to effectively prepare data for training machine learning models. Additionally, the library provides tools for data splitting, and model evaluation to ensure reliable and robust model performance assessment.

Clinical Applications of Machine Learning.

This review introduces interpretable predictive machine learning approaches, natural language processing, image recognition, and reinforcement learning methodologies to familiarize end users. As machine learning, artificial intelligence, and generative artificial intelligence become increasingly utilized in clinical medicine, it is imperative that end users understand the underlying methodologies. This review describes publicly available datasets that can be used with interpretable predictive approaches, natural language processing, image recognition, and reinforcement learning models, outlines result interpretation, and provides references for in-depth information about each analytical framework. This review introduces interpretable predictive machine learning models, natural language processing, image recognition, and reinforcement learning methodologies. Interpretable predictive machine learning models, natural language processing, image recognition, and reinforcement learning are core machine learning methodologies that underlie many of the artificial intelligence methodologies that will drive the future of clinical medicine and surgery. End users must be well versed in the strengths and weaknesses of these tools as they are applied to patient care now and in the future.

Creating data-driven products and services in industry 4.0: a case study on companies in the German machine and tool industry

During the past ten years, lots of new data-driven products and services for tools, machinery and equipment have been developed. While several new players from other industries gained a certain market share, plant and machinery producers also started to enhance their portfolio to take on new data-driven products and services because of the technological changes in Industry 4.0. As a first part of the research, an extensive market study was carried out to analyze how many German companies already offer data-based products and services in addition to their core machines and understand what kind of offerings they make. To classify these offerings, a scheme based on established Industry 4.0 maturity models was developed. In brief, the market for data-driven products and services is still developing, with few technology leaders and fast movers taking the largest share. While the market study gave an overview of what was on offer, the second part of this contribution analyzes how the fast movers with a high level of Industry 4.0 maturity conducted their data-driven services and products. Thus, these few companies were analyzed in more detail, based on public material as well as subsequent expert interviews. Most fast movers in this study relied on the same patterns and approaches, especially when looking at organizational issues such as customer-driven innovation, agile organization of operations, mixed teams, partnering and portfolio enhancement.

WiRITE: General and Practical Wi-Fi Based Hand-Writing Recognition

Device-free hand-writing systems identify the content that a user writes by hand movement in the air, thus providing an intuitive human computer interface. In this paper, we propose WiRITE, a Wi-Fi hand-writing recognition system built with commodity Wi-Fi APs. Unlike most existing machine learning based hand-writing recognition systems, which are often subject to severe limitations in generality, e.g., high training overhead when adapted across hand-writing alphabets, environments, and users, WiRITE is designed with unique consideration of its generality when applied to practice—being application-transferable, environment-agnostic, and user-independent. With little training overhead, WiRITE behaves inclusively to different users, environments, and applications, stemming from a comprehensive design of signal processing that is built into its core machine learning model. Extensive evaluation is conducted with five users for three applications, i.e., recognizing Digits, English letters, and Chinese characters, in realistic office environment. The experiment results demonstrate that WiRITE provides at least 0.9 accuracy in various combinations of users and applications with 0.93 accuracy in average.

Integrating core physics and machine learning for improved parameter prediction in boiling water reactor operations

This study introduces a novel method for enhancing Boiling Water Reactor (BWR) operation simulations by integrating machine learning (ML) models with conventional simulation techniques. The ML model is trained to identify and correct errors in low-fidelity simulation outputs, traditionally derived from core physics computations. These corrections aim to align the low-fidelity results closely with high-fidelity data. Precise predictions of nuclear reactor parameters like core eigenvalue and power distribution are crucial for efficient fuel management and adherence to technical specifications. Current high-fidelity transport calculations, while accurate, are impractical for real-time predictions due to extensive computational demands. Our approach, therefore, utilizes the standard two-step simulation process-assembly-level lattice physics calculations followed by whole-core nodal diffusion computations-to generate initial results, which are then refined using the ML-based error correction model. The methodology focuses on improving simulation accuracy in regular BWR operations rather than developing a universal ML predictor for reactor physics. By training an advanced neural network model on the difference in high-fidelity and low-fidelity simulations, the model can reduce the nodal power error from low-fidelity simulations to around 1% on average and the core eigenvalue down to under 100 pcm. This result is under the condition of the normal variations of control rod pattern and core flow rate changes in standard BWR operations used in the training and evaluation of the machine learning model. This work suggests a promising approach for achieving more accurate, computationally feasible simulation solutions in nuclear reactor operation and management.

Heat transfer analysis in particle-laden flows using the immersed boundary method

This paper investigates an efficient immersed boundary method (IBM) for multiple-core CPU machines with local grid refinement for the calculation of heat transfer between fluids and finite-sized particles. The fluid momentum equations are solved by using the fractional step method, while the energy equation is solved by employing the second-order Adams-Bashforth method. For efficient load balancing between the CPU cores, the coupling between particles and fluid is obtained by applying the body force in the fluid equations, which depends on the solid volume fraction of particles contained in each grid cell, and then by linearly interpolating the particle temperature and velocity on the fluid grid cell (in place of the delta function commonly used in literature). Several test cases from the literature are studied, and good agreement is observed between the simulation results and the literature. Finally, a scaling study on multiple core machines is performed, demonstrating the proposed IBM's capabilities for a significant reduction in processing time.

Quench detection electronics for FAIR

The SIS100 synchrotron and the Super-FRS fragment separator are the core machines of the Facility for Antiproton and Ion Research (FAIR), currently under construction in Darmstadt, Germany. A uniform quench detection system will ensure monitoring of superconducting magnets of both SIS100 and Super-FRS. Further, Plasma Physics and Applications (APPA) experiments will also make use of it.The distributed quench detection system of FAIR is based on hundreds of quench detection units. Their design, testing experience and first operational results are presented in this work.

Digital Simulation and Analysis of Assembly-Deviation Prediction Based on Measurement Data

For various power equipment items such as aircraft engines and gas turbines with numerous components demanding requirements on processing accuracy and high structural complexity, processing errors tend to cause such problems as assembly rework, forced assembly and prolonged research and development cycle. In this paper, the core machine of micro gas turbine is taken as the research object to construct a simulation model of assembly-deviation prediction based on the design tolerance and actual measurement data. Then, an analysis is conducted on the assemblability of the design model and the key factors causing the deviation of the assembly. After conducting deviation calculations and simulation analysis, it was determined that the current mounting position is deemed to be suboptimal. In light of this finding, an optimized solution is proposed, which involves advancing the mounting phase by 0.47 mm. This adjustment effectively resolves the interference issue resulting from the design error. Moreover, the geometry, shape and three-dimensional contour of the key components are measured with high accuracy and precision to identify the key characteristic parameters affecting the outcome of the assembly. With an assembly-deviation-prediction and -analysis model established on the basis of actual measurement data, the results of the assembly-deviation analysis are compared with the outcome of the assembly, showing high consistency. The assembly-deviation-prediction method proposed in this paper on the basis of design tolerance and actual measurement data is applicable to the manufacture of aviation and combustion engines.

Efficient multivariate optimization of an ultrasonic transducer with genetic parallel algorithms

Design optimization of a multilayer transducer is desired for flaw detection, biomedical and underwater applications by controlling layer thicknesses of active material and volume fractions of ceramic in each piezocomposite layer. To handle the computational complexity in solving such multivariate optimization and the complex nature of the problem, we propose a parallel genetic algorithm (GA)-based strategy. The complexity of the problem increases with the number of layers and layer-wise ceramic volume fraction to achieve an optimized transducer working efficiently in an operational frequency range as desired by the application specification. The fitness function is formulated based on the transducer response in terms of about the same value of pressure magnitude at the first three harmonics at functional frequencies. The fitness value is computed using a one-dimensional model (ODM) for PMNPT, PZT4d, and PZT5h-based materials initially, while input layer thicknesses and volume fractions are evaluated through GAs and passed to ODM. The overall scheme is maintained through parallel GAs. The simulation is performed by using 1, 4, 6, and 8 core machines to reduce time complexity. The results for parallel GA-ODMs are statistically represented in the minimum, maximum, mean and standard deviation of fitness, while graphically represented for speedup and time.

Seismic urban damage map generation based on satellite images and Gabor convolutional neural networks

Rapid assessment of urban damages after a strong earthquake is a necessary and crucial task to reduce the number of fatalities and recover socioeconomic services. In this paper, a novel deep-learning-based framework is proposed for detecting and mapping damages in urban buildings and roads using post-earthquake high-resolution satellite imagery. The method begins with overlaying a pre-event vector map on an input image to extract the building and road objects. The core machine learning components include two separate convolutional neural networks (CNN), integrated with Gabor filters, which extract debris pixels associated with building and road objects. These debris pixels are analyzed to generate the final damage maps, which show multiple damage degrees for buildings and roads. Two different datasets were used to thoroughly evaluate the proposed method's overall effectiveness. The overall accuracy of 95% for detecting the debris pixels in building and road areas proves the effectiveness of the proposed CNN models for debris detection in comparison to the traditional Machine Learning (ML) methods. The proposed method successfully labelled 84% of the buildings and 87% of the roads when compared with a manually generated multiple damage map.

Expanding the Diversity of Accumulibacter with a Novel Type and Deciphering the Transcriptional and Morphological Features among Co-Occurring Strains.

"Candidatus Accumulibacter" is the major polyphosphate-accumulating organism (PAO) in global wastewater treatment systems, and its phylogenetic and functional diversity have expanded in recent years. In addition to the widely recognized type I and II sublineages, we discovered a novel type enriched in laboratory bioreactors. Core gene and machine learning-based gene feature profiling supported the assertion that type III "Ca. Accumulibacter" is a potential PAO with the unique function of using dimethyl sulfoxide as an electron acceptor. Based on the correlation between ppk1 and genome similarity, the species-level richness of Accumulibacter was estimated to be over 100, suggesting that the currently recognized species are only the tip of the iceberg. Meanwhile, the interstrain transcriptional and morphological features of multiple "Ca. Accumulibacter" strains co-occurring in a bioreactor were investigated. Metatranscriptomics of seven co-occurring strains indicated that the expression level and interphasic dynamics of PAO phenotype-related genes had minimal correlation with their phylogeny. In particular, the expression of denitrifying and polyphosphate (poly-P) metabolism genes exhibited higher interstrain and interphasic divergence than expression of glycogen and polyhydroxyalkanoate metabolic genes. A strategy of cloning rRNA genes from different strains based on similar genomic synteny was successfully applied to differentiate their morphology via fluorescence in situ hybridization. Our study further expands the phylogenetic and functional diversity of "Ca. Accumulibacter" and proposes that deciphering the function and capability of certain "Ca. Accumulibacter" should be tailored to the environment and population in question. IMPORTANCE In the last 2 decades, "Ca. Accumulibacter" has garnered significant attention as the core functional but uncultured taxon for enhanced biological phosphorus removal due to its phylogenetic and functional diversity and intragenus niche differentiation. Since 2002, it has been widely known that this genus has two sublineages (type I and II). However, in this study, a metagenomic approach led to the discovery of a novel type (type III) with proposed novel functional features. By comparing the average nucleotide identity of "Ca. Accumulibacter" genomes and the similarity of ppk1, a phylogenetic biomarker largely deposited in databases, the global species-level richness of "Ca. Accumulibacter" was estimated for the first time to be over 100. Furthermore, we observed the co-occurrence of multiple "Ca. Accumulibacter" strains in a single bioreactor and found the simultaneous transcriptional divergence of these strains intriguing with regard to their niche differentiation within a single community. Our results indicated a decoupling feature between transcriptional pattern and phylogeny for co-occurring strains.

High-resolution core data and machine learning schemes applied to rock facies classification

Abstract This paper presents core analysis practices developed to increase the value of cores in reservoir characterization workflows. Core analysis traditionally implies extensive rock testing programmes that require large numbers of plug samples. Numerous stakeholders compete for intact core material and often do not base sample selection criteria on objective and reliable information. As a consequence, samples dedicated to core analysis programmes consume a significant fraction of the material available, yet they are selected on the basis of very little a priori information and therefore do not necessarily capture the complexity of the targeted formations. In an attempt to change this paradigm, we promote a more integrated core analysis solution combining transdisciplinary, high-resolution, non-destructive measurements on whole cores, for an early yet objective description of cores and the rapid estimation of formation properties. The first section describes multidisciplinary and non-destructive tests designed to increase the value of information from cores while keeping a minimal footprint. Core samples are scanned along a fit-for-purpose surface, with a collection of sensors interfaced to the same table-top equipment. Technologies including ultra-high-resolution pictures, elemental composition and the direct measurements of geomechanical properties provide continuous and high-resolution profiles with a unique depth reference. Panoramic pictures are processed to extract textural and colour features and grain size distributions. Grain size distributions are backed up by analysis of topographical maps created with a laser scan. The results are combined under one unique format, thereby easing interdisciplinary work from the verification of standard tests (routine core analysis, rock mechanical test) to the construction of robust predictive rock models. The paper describes machine learning schemes applied to core datasets. Unsupervised schemes are designed to identify rock facies, while supervised schemes are used to classify tested rocks into predefined rock facies with known petrophysical properties. Case studies highlight the benefits of this approach of core analysis in conjunction with artificial intelligence for the automated identification and classification of rock facies. This novel approach to core analysis leverages a detailed and comprehensive knowledge of the distributions of core properties, available under one unique format for all disciplines, which eases interdisciplinary work and significantly improves existing core analysis standards. It also provides a sound basis to train artificial intelligence rock reservoir property predictors linking well-log data to core-based lithofacies signatures.

Insights into azalomycin F assembly-line contribute to evolution-guided polyketide synthase engineering and identification of intermodular recognition

Modular polyketide synthase (PKS) is an ingenious core machine that catalyzes abundant polyketides in nature. Exploring interactions among modules in PKS is very important for understanding the overall biosynthetic process and for engineering PKS assembly-lines. Here, we show that intermodular recognition between the enoylreductase domain ER1/2 inside module 1/2 and the ketosynthase domain KS3 inside module 3 is required for the cross-module enoylreduction in azalomycin F (AZL) biosynthesis. We also show that KS4 of module 4 acts as a gatekeeper facilitating cross-module enoylreduction. Additionally, evidence is provided that module 3 and module 6 in the AZL PKS are evolutionarily homologous, which makes evolution-oriented PKS engineering possible. These results reveal intermodular recognition, furthering understanding of the mechanism of the PKS assembly-line, thus providing different insights into PKS engineering. This also reveals that gene duplication/conversion and subsequent combinations may be a neofunctionalization process in modular PKS assembly-lines, hence providing a different case for supporting the investigation of modular PKS evolution.

Auto JSON: An Automatic Transformation Model for Converting Relational Database to Non-relational Documents

In recent days, the demand for dealing large set of distributed data obsoletes the relational database and its structured query language (SQL) solutions in practice and paves the way for novel solutions in the name of non-relational database as not-only SQL (NoSQL). The NoSQL offers dynamic, flexible, scalable, highly available, greater performance and near real-time access to the distributed nature of voluminous data used for current industrial applications. Apart from these giant features of NoSQL, the SQL is still found to be in operation because of its popularity and standard. This paper projected an algorithm to convert the relational documents of MySQL into any document oriented NoSQL databases automatically without destructing the existing relational database setup and installing the NoSQL from scratch in the core machines. Java Script Object Notation (JSON) is a human readable data interchange format, being used in web development. The characteristics of JSON widened its use cases from web development to database storage. The Mongo DB, one of the most popular document oriented NoSQL adapts JSON format for its storage. The proposed algorithm is built based on its schema definition and the performance is captured through evaluating it against a sample database from hospital management system. The findings are discussed with great interest of addressing the challenges and revealing the scope for improvement.

Scope of repurposed drugs against the potential targets of the latest variants of SARS-CoV-2.

The unprecedented outbreak of the severe acute respiratory syndrome (SARS) Coronavirus-2, across the globe, triggered a worldwide uproar in the search for immediate treatment strategies. With no specific drug and not much data available, alternative approaches such as drug repurposing came to the limelight. To date, extensive research on the repositioning of drugs has led to the identification of numerous drugs against various important protein targets of the coronavirus strains, with hopes of the drugs working against the major variants of concerns (alpha, beta, gamma, delta, omicron) of the virus. Advancements in computational sciences have led to improved scope of repurposing via techniques such as structure-based approaches including molecular docking, molecular dynamic simulations and quantitative structure activity relationships, network-based approaches, and artificial intelligence-based approaches with other core machine and deep learning algorithms. This review highlights the various approaches to repurposing drugs from a computational biological perspective, with various mechanisms of action of the drugs against some of the major protein targets of SARS-CoV-2. Additionally, clinical trials data on potential COVID-19 repurposed drugs are also highlighted with stress on the major SARS-CoV-2 targets and the structural effect of variants on these targets. The interaction modelling of some important repurposed drugs has also been elucidated. Furthermore, the merits and demerits of drug repurposing are also discussed, with a focus on the scope and applications of the latest advancements in repurposing.

Development of a predictive model for university students’ core competency index using machine learning: Focusing on D University

Objectives The purpose of this study was to present implications for the development and operation of the core competency-based curriculum by predicting the core competency of college students.
 Methods The panel data of the D-CODA result panel data for the past 3 years (2019-2021) of D-University students located in the Busan area were analyzed. Machine learning prediction models such as multiple linear regression analysis (LR), random forest (RF), and support vector machine (SVM), were used to predict core competencies.
 Results The following research results were derived from the study. First, the optimal prediction model for each core competency is as follows. Professional competency was shown in the RF (random forest) model, personality competency in SVM (support vector machine), creative competency in the RF (random forest) model, challenge competency and glocal (global and local) competency in the SVM (support vector machine) model, and communication competency in the LR (multi-linear regression analysis) model. Second, in the analysis of competencies, it was found that professional competency contributes to the prediction of professional competency, and both personality competency and communication competency to that of personality competency. Third, in the model analysis to predict the overall core competency index, the optimal predictive model was found to be the RF (random forest) model which showed the least error. Fourth, in the prediction of key competency indicators in 2022, it is predicted that expertise, personality, creativity, and challenging competency will improve.
 Conclusions This study revealed that the analysis using accumulated core competency data and machine learning is useful in predicting and discriminating the core competency of college students. This study is meaningful in that it suggests the importance of periodic core competency index management at the university level and provides the basis for designing a core competency-based curriculum.

A Data-Driven Lens to Understand Human Biology: An Interview with Daphne Koller

A Data-Driven Lens to Understand Human Biology: An Interview with Daphne Koller

Predictive Coding Approximates Backprop Along Arbitrary Computation Graphs.

Backpropagation of error (backprop) is a powerful algorithm for training machine learning architectures through end-to-end differentiation. Recently it has been shown that backprop in multilayer perceptrons (MLPs) can be approximated using predictive coding, a biologically plausible process theory of cortical computation that relies solely on local and Hebbian updates. The power of backprop, however, lies not in its instantiation in MLPs but in the concept of automatic differentiation, which allows for the optimization of any differentiable program expressed as a computation graph. Here, we demonstrate that predictive coding converges asymptotically (and in practice, rapidly) to exact backprop gradients on arbitrary computation graphs using only local learning rules. We apply this result to develop a straightforward strategy to translate core machine learning architectures into their predictive coding equivalents. We construct predictive coding convolutional neural networks, recurrent neural networks, and the more complex long short-term memory, which include a nonlayer-like branching internal graph structure and multiplicative interactions. Our models perform equivalently to backprop on challenging machine learning benchmarks while using only local and (mostly) Hebbian plasticity. Our method raises the potential that standard machine learning algorithms could in principle be directly implemented in neural circuitry and may also contribute to the development of completely distributed neuromorphic architectures.

Accelerating Edit-Distance Sequence Alignment on GPU Using the Wavefront Algorithm

Sequence alignment remains a fundamental problem with practical applications ranging from pattern recognition to computational biology. Traditional algorithms based on dynamic programming are hard to parallelize, require significant amounts of memory, and fail to scale for large inputs. This work presents eWFA-GPU, a GPU (graphics processing unit)-accelerated tool to compute the exact edit-distance sequence alignment based on the wavefront alignment algorithm (WFA). This approach exploits the similarities between the input sequences to accelerate the alignment process while requiring less memory than other algorithms. Our implementation takes full advantage of the massive parallel capabilities of modern GPUs to accelerate the alignment process. In addition, we propose a succinct representation of the alignment data that successfully reduces the overall amount of memory required, allowing the exploitation of the fast shared memory of a GPU. Our results show that our GPU implementation outperforms by 3-9× the baseline edit-distance WFA implementation running on a 20 core machine. As a result, eWFA-GPU is up to 265 times faster than state-of-the-art CPU implementation, and up to 56 times faster than state-of-the-art GPU implementations.

Machine Learning in Precision Agriculture: A Survey on Trends, Applications and Evaluations Over Two Decades

Precision agriculture represents the new age of conventional agriculture. This is made possible by the advancement of various modern technologies such as the internet of things. The unparalleled potential for data collection and analytics has resulted in an increase in multi-disciplinary research within machine learning and agriculture. However, the application of machine learning techniques to agriculture seems to be out of step with core machine learning research. This gap is further exacerbated by the inherent challenges associated with agricultural data. In this work, we conduct a systematic review of a large body of academic literature published between 2000 and 2022, on the application of machine learning techniques to agriculture. We identify and discuss some of the key data issues such as class imbalance, data sparsity and high dimensionality. Further, we study the impact of these data issues on various machine learning approaches within the context of agriculture. Finally, we identify some of the common pitfalls in the machine learning and agriculture research including the misapplication of machine learning evaluation techniques. To this end, this survey presents a holistic view on the state of affairs in the cross-domain of machine learning and agriculture and proposes some suitable mitigation strategies to address these challenges.