Domain Knowledge Research Articles

Towards Model-Driven Explainable Artificial Intelligence: Function Identification with Grammatical Evolution

Machine learning is a well-matured discipline, and exploration of datasets can be performed in an efficient way, leading to accurate and operational prediction and decision models. On the other hand, most methods tend to produce black-box-type models, which can be considered a serious drawback. This is so, since in case of numerous practical applications, it is also required to justify, explain, and uncover the inner decision mechanism so that an in-depth understanding of the causal and functional dependencies becomes possible and some responsibility for the decision can be considered. This paper addresses the critical need for model-driven eXplainable Artificial Intelligence (XAI) by exploring the limitations inherent in existing explanatory mechanisms, such as LIME or SHAP, which rely solely on input data. This seems to be an intrinsic limitation and a conceptual error, as no expert domain knowledge can come into play, and no analytical models of the phenomena under investigation are created. In order to deal with this issue, this paper puts forward the idea of building open, white-box explanatory models. To do that, we propose employing grammatical evolution tools combined with expert domain knowledge. The results demonstrate that the developed models can effectively explain the structure and behavior of decision models in terms of components, connections, causality, and simple functional dependencies.

Multi-physics information-integrated neural network for fatigue life prediction of additively manufactured Hastelloy X superalloy

ABSTRACT The inherent ‘black-box’ characteristic of neural networks renders the physical interpretability of fatigue life prediction challenging, resulting in physically inconsistent prediction results. In response to this challenge, a multi-physics information-integrated neural network framework is proposed for fatigue life prediction. Rooted in the continuous damage mechanics with embedded additive manufacturing process parameters (AM-CDM), coupled with activation functions and loss functions that amalgamate physical information, a multi-tiered and multi-physical information source network framework emerges. The framework incorporates the outcomes of the AM-CDM model as input features for domain knowledge enhancement, embeds the fatigue limit within the activation function, and establishes a loss function integrating the physical model. Through the integration of multiple physical information, the model can be restricted by different physical constraints during training, thereby improving its prediction accuracy and physical consistency. Validation of the proposed framework is conducted using laser powder bed fusion (LPBF)-fabricated Hastelloy X subjected to varying orientations, post-treatments and test temperatures. The findings demonstrate that the prediction accuracy of an integrating-physics-informed neural network (IPINN) is improved by 27.51%, and its life prediction results are capped at the fatigue limit life, yielding a better physical consistency. The IPINN framework provides a novel perspective for fatigue life prediction.

Predicting the Liquid Steel End-Point Temperature during the Vacuum Tank Degassing Process Using Machine Learning Modeling

The present work focuses on predicting the steel melt temperature following the vacuum treatment step in a vacuum tank degasser (VTD). The primary objective is to establish a comprehensive methodology for developing and validating machine learning (ML) models within this context. Another objective is to evaluate the model by analyzing the alignment of the SHAP values with metallurgical domain expectations, thereby validating the model’s predictions from a metallurgical perspective. The proposed methodology employs a Random Forest model, incorporating a grid search with domain-informed variables grouped into batches, and a robust model-selection criterion that ensures optimal predictive performance, while keeping the model as simple and stable as possible. Furthermore, the Shapley Additive Explanations (SHAP) algorithm is employed to interpret the model’s predictions. The selected model achieved a mean adjusted R2 of 0.631 and a hit ratio of 75.3% for a prediction error within ±5 °C. Despite the moderate predictive performance, SHAP highlighted several aspects consistent with metallurgical domain expertise, emphasizing the importance of domain knowledge in interpreting ML models. Improving data quality and refining the model framework could enhance predictive performance.

Development of self-derivation through memory integration and relations with world knowledge

ABSTRACT Accumulating world knowledge is a major task of development and education. The productive process of self-derivation through memory integration seemingly is a valid model of the process. To test the model, we examined relations between generation and retention of new factual knowledge via self-derivation through integration and world knowledge as measured by standardised assessments. We also tested whether the productive process of self-derivation predicted world knowledge even when a measure of learning through direct instruction also was considered. Participants were 162 children ages 8–12 years (53% female; 15% Black, 6% Asian, 1% Arab, 66% White, 5% mixed race, 7% unreported; 1% Latinx). Age accounted for a maximum of 4% of variance in self-derivation and retention. In contrast, substantial individual variability related to general knowledge and content knowledge in several domains, explaining 20–40% variance. In each domain for which self-derivation performance was a unique predictor, it explained a nominally greater share of the variance than the measure of learning through direct instruction. The findings imply that individual variability in self-derivation has functional consequences for accumulation of semantic knowledge across the elementary-school years.

Context-aware reinforcement learning for cooling operation of data centers with an Aquifer Thermal Energy Storage

Data centers are often equipped with multiple cooling units. Here, an aquifer thermal energy storage (ATES) system has shown to be efficient. However, the usage of hot and cold-water wells in the ATES must be balanced for legal and environmental reasons. Reinforcement Learning has been proven to be a useful tool for optimizing the cooling operation at data centers. Nonetheless, since cooling demand changes continuously, balancing the ATES usage on a yearly basis imposes an additional challenge in the form of a delayed reward. To overcome this, we formulate a return decomposition, Cool-RUDDER, which relies on simple domain knowledge and needs no training. We trained a proximal policy optimization agent to keep server temperatures steady while minimizing operational costs. Comparing the Cool-RUDDER reward signal to other ATES-associated rewards, all models kept the server temperatures steady at around 30 °C. An optimal ATES balance was defined to be 0% and a yearly imbalance of −4.9% with a confidence interval of [−6.2, −3.8]% was achieved for the Cool 2.0 reward. This outperformed a baseline ATES-associated reward of 0 at −16.3% with a confidence interval of [−17.1, −15.4]% and all other ATES-associated rewards. However, the improved ATES balance comes with a higher energy consumption cost of 12.5% when comparing the relative cost of the Cool 2.0 reward to the zero reward, resulting in a trade-off. Moreover, the method comes with limited requirements and is applicable to any long-term problem satisfying a linear state-transition system.

Frequency-focused sound data generator for fault diagnosis in industrial robots

Abstract A frequency-focused sound data generator was developed for the in situ fault sound diagnosis of industrial robot reducers. The sound data generator, based on a conditional generative adversarial network, selects a target frequency range without relying on domain knowledge. A sound dataset of normal and faulty harmonic drive rotations of in situ industrial robots was collected using an attachable wireless sound sensor. The generated sound data were evaluated based on the fault diagnosis accuracy of a simple classifier trained using the generated data and tested using real data. The proposed method well-defined the frequency feature clusters and produced high-quality data, exhibiting up to 16.0% higher precision score on normal and 13.0% higher accuracy on weak-fault harmonic drive compared with the conventional methods, achieving fault diagnosis accuracy of 95.6% even in situations of fault data comprising only 5% of the normal data.

Emerging Nanozymes in Neurological Disorder Therapeutics: Bridging Oxidoreductase Mimicry and Antioxidant Chemistry

AbstractThe prevalence of neurological dieases, including neurodegenerative, neurotraumatic disorders, and neuroinflammatory conditions, has been rising due to global population and aging demographics. A key factor in the pathogenesis of these disorders is the hyperaccumulation of reactive oxygen and nitrogen species (RONS). Nanozymes have emerged as promising candidates for neurotherapeutic applications owing to their exceptional catalytic activity and stability. Of particular note is their ability to cross the blood‐brain barrier and counteract the production of reactive oxygen species via their enzyme‐mimicking characteristics. In this review, the latest advancements and theoretical knowledge in this research domain are summarized. Using the inherent functionalities of the Web of Science and bibliometric methodologies, annual publication trends are identified and extensively explored the most researched topics and neurological disorders in this field. The antioxidant reduction chemistry of the nanozymes is discussed, highlighting their ability to mimic natural oxidoreductase activity and inhibit RONS production at the source. Moreover, this review delves into the current limitations and future prospects of these mechanisms in addressing neurological disorders. The significant benefits and recent developments in the use of RONS‐regulating nanozymes for the treatment of neurological diseases are emphasized, offering insights into their therapeutic applications and broader implications for neurology.

Women Empowerment through Self Help Groups: A Bibliometric Analysis

‘Women Empowerment’ has been regarded as an inevitable component of development. The Self Help Groups (SHGs) approach was implemented to promote savings and create social cohesiveness among women to empower them in a collective manner, especially for the rural poor. The article conducted a comprehensive bibliometric analysis to identify significant contributors, existing refinements, fields, and advocates for different directions in the study area of SHGs and women's empowerment and to ascertain the domain knowledge on developments connected to women empowerment and SHGs. The Scopus database was used to search the global literature on self-help groups and Women Empowerment. As per the findings, this field is still growing globally, both in developed and developing nations. It found significant links between SHGs and socio-economic empowerment, gender equality, and health promotion. It also shows that SHGs have been utilized as a medium at the local, national, and worldwide levels to achieve gender equality and women's empowerment in various countries.

Cognitive intelligence-enabled requirements elicitation for design optimisation of smart product-service system

The rapid development of advanced technologies provides futuristic option to enhance product value with digital services, which the concept of Smart Product-Service System (SPSS) was proposed. SPSS focuses on usage data, the whole lifecycle, and the iterative optimisation. A large amount of data and context information can be collected during the usage stage, which can be used to elicit requirements to guide further design. How to realise the requirement elicitation with cognitive intelligence based on usage data is the key step for design optimisation. Therefore, this research proposes a cognitive intelligence-enabled requirements elicitation framework to help designers promote the design optimisation process. First, the digital twin-driven data collection architecture of SPSS is proposed by considering multi-sources and multi-stages. Second, the behaviour of SPSS is modelled based on Auto-Encoder which is constructed using deep learning to cognise SPSS performance. Third, the corresponding behaviour model is interpreted based on Shapley Addictive exPlanations (SHAP) to discover key factors with cognitive intelligence. Finally, the domain knowledge graph is used to generate the optimisation requirements for further design. To demonstrate the feasibility and advantages of the proposed framework, this study adopts the case study of wind turbines for illustration and verification.

Ensemble of HMMs for Sequence Prediction on Multivariate Biomedical Data

Background: Biomedical data are usually collections of longitudinal data assessed at certain points in time. Clinical observations assess the presences and severity of symptoms, which are the basis for the description and modeling of disease progression. Deciphering potential underlying unknowns from the distinct observation would substantially improve the understanding of pathological cascades. Hidden Markov Models (HMMs) have been successfully applied to the processing of possibly noisy continuous signals. We apply ensembles of HMMs to categorically distributed multivariate time series data, leaving space for expert domain knowledge in the prediction process. Methods: We use an ensemble of HMMs to predict the loss of free walking ability as one major clinical deterioration in the most common autosomal dominantly inherited ataxia disorder worldwide. Results: We present a prediction pipeline that processes data paired with a configuration file, enabling us to train, validate and query an ensemble of HMMs. In particular, we provide a theoretical and practical framework for multivariate time-series inference based on HMMs that includes constructing multiple HMMs, each to predict a particular observable variable. Our analysis is conducted on pseudo-data, but also on biomedical data based on Spinocerebellar ataxia type 3 disease. Conclusions: We find that the model shows promising results for the data we tested. The strength of this approach is that HMMs are well understood, probabilistic and interpretable models, setting it apart from most Deep Learning approaches. We publish all code and evaluation pseudo-data in an open-source repository.

Evaluation of artificial intelligence-generated drug therapy communication skill competencies in medical education.

This study compared three artificial intelligence (AI) platforms' potential to identify drug therapy communication competencies expected of a graduating medical doctor. We presented three AI platforms, namely, Poe Assistant©, ChatGPT© and Google Bard©, with structured queries to generate communication skill competencies and case scenarios appropriate for graduating medical doctors. These case scenarios comprised 15 prototypical medical conditions that required drug prescriptions. Two authors independently evaluated the AI-enhanced clinical encounters, which integrated a diverse range of information to create patient-centred care plans. Through a consensus-based approach using a checklist, the communication components generated for each scenario were assessed. The instructions and warnings provided for each case scenario were evaluated by referencing the British National Formulary. AI platforms demonstrated overlap in competency domains generated, albeit with variations in wording. The domains of knowledge (basic and clinical pharmacology, prescribing, communication and drug safety) were unanimously recognized by all platforms. A broad consensus among Poe Assistant© and ChatGPT© on drug therapy-related communication issues specific to each case scenario was evident. The consensus primarily encompassed salutation, generic drug prescribed, treatment goals and follow-up schedules. Differences were observed in patient instruction clarity, listed side effects, warnings and patient empowerment. Google Bard did not provide guidance on patient communication issues. AI platforms recognized competencies with variations in how these were stated. Poe Assistant© and ChatGPT© exhibited alignment of communication issues. However, significant discrepancies were observed in specific skill components, indicating the necessity of human intervention to critically evaluate AI-generated outputs.

Informating and Automating in Systems Design: Extending the Zuboffian View of Technology Ready-to-Use vs. Technology Unintuitive-to-Use

ABSTRACT Previous literature has treated software functionalities as a binary choice between informating and automating, based on Zuboff’s definitions. We suggest that the space between the role of a user engaging with a system (informate) and what the system can accomplish for the user (automate) has increasingly become blurred, moving toward a more nuanced continuum. We present a user-system interaction framework that analyzes this continuum with a focus on the user’s perspective. Using cases from archival data, this paper provides an analysis of what goes well and badly in the locus of control handoff between users and systems, highlighting the implications of software that is created in one region but used globally. The paper contributes to the literature by extending Zuboff’s informate vs. automate discussion and identifying design decisions that could help to avoid potentially catastrophic design errors. Using the interaction framework, we argue that it would be beneficial for system designers to move toward the informate end of the continuum, when designing systems for users with a high level of domain knowledge and expertise, and when there are significant differences between the environment in which the system is designed and the one in which the system is used.

A Hierarchical Graph Neural Network Framework for Predicting Protein-Protein Interaction Modulators With Functional Group Information and Hypergraph Structure.

Accurate prediction of small molecule modulators targeting protein-protein interactions (PPIMs) remains a significant challenge in drug discovery. Existing machine learning-based models rely on manual feature engineering, which is tedious and task-specific. Recently, deep learning models based on graph neural networks have made remarkable progress in molecular representation learning. However, many graph-based approaches ignore molecular hierarchical structure modeling guided by domain knowledge. In chemistry, the functional groups of a molecule determine its interaction with specific targets. Therefore, we propose a hierarchical graph neural network framework (called HiGPPIM) for predicting PPIMs by integrating atom-level and functional group-level features of molecules. HiGPPIM constructs atom-level and functional group-level graphs based on chemical knowledge and learns graph representations using graph attention networks. Furthermore, a hypergraph attention network is designed in HiGPPIM to aggregate and transform two-level graph information. We evaluate the performance of HiGPPIM on eight PPI families and two prediction tasks, namely PPIM identification and potency prediction. Experimental results demonstrate that HiGPPIM achieves state-of-the-art performance on both tasks and that using functional group information to guide PPIM prediction is effective.

Multi-objective optimization of multi-principal element alloys via high-throughput simulation and active learning

It is a great challenge to optimize the composition and improve the multi-properties of multi-principal element alloys (MPEAs) by traditional trial-and-error method. Here, an active learning strategy is proposed to design the multi-objective properties of MPEAs based on molecular dynamics (MD) simulation, machine learning (ML) models and multi-objective optimization algorithms. A database is established via a tensile test of 120 FeCrNiCoMn single-crystal samples by MD simulation. Six ML models are employed to predict the yield strength and Young's modulus based on domain knowledge, among which the support vector machine (SVM) model shows the better predictive performance, with low mean absolute error of 0.12 and 0.61 for cross validation, respectively. Then, by combining the SVM model and three multi-objective optimization algorithms, optimal alloy compositions are rapidly searched in a virtual composition space containing 885681 alloys. After four optimization-testing iterations, Co35Cr30Ni35 is synthesized with simultaneous high yield strength and Young's modulus, which are 2.9 and 1.6 times as equiatomic FeCrNiCoMn alloy, respectively.

Exploration of deep-learning-based dose rate estimation model for a large-volume plastic scintillation detector

This study explores a deep-learning-based dose rate estimation model for a large-volume plastic scintillation detector. Large-volume plastic scintillators are typically utilized as detector materials for radiation portal monitors (RPMs); however, they can also be utilized for environmental radiation monitoring. We introduced a deep-learning approach to estimate the ambient dose equivalent (H*(10)) from the spectral input using a transformer architecture. The performance of the model was compared with that of other models with conventional architectures and verified using simulated and measured spectra. In addition, we present a model explanation and an uncertainty quantification approach to ensure the reliability of the dose rate estimation model. Our verification demonstrates that the behavior of the model aligns with domain knowledge and that the uncertainty of the model's estimations increases according to the increment in the statistical uncertainties of the input spectra. The experimental results indicate that the model shows a promising performance as a practical application for RPMs.

Automatic Metric Search for Few-Shot Learning.

Few-shot learning (FSL) aims to learn a model that can identify unseen classes using only a few training samples from each class. Most of the existing FSL methods adopt a manually predefined metric function to measure the relationship between a sample and a class, which usually require tremendous efforts and domain knowledge. In contrast, we propose a novel model called automatic metric search (Auto-MS), in which an Auto-MS space is designed for automatically searching task-specific metric functions. This allows us to further develop a new searching strategy to facilitate automated FSL. More specifically, by incorporating the episode-training mechanism into the bilevel search strategy, the proposed search strategy can effectively optimize the network weights and structural parameters of the few-shot model. Extensive experiments on the miniImageNet and tieredImageNet datasets demonstrate that the proposed Auto-MS achieves superior performance in FSL problems.

Tax aggressiveness in a scenario of economic crisis and financial difficulty: an analysis of companies listed on B3

Objective: This study aimed to examine the relationship between economic crisis scenarios, financial difficulties, and tax aggressiveness in companies listed on B3. Methodology: This descriptive and documentary study employed a quantitative approach, incorporating a sample of 326 companies. Data analysis entailed using descriptive statistics and regression analysis with panel data. Results: The findings of this study suggest that a crisis environment does not promote the adoption of aggressive tax strategies by the companies within the sample. This contradicts established literature, which has posited that there is a heightened effort to decrease tax liability during periods of crisis. Nonetheless, it was observed that companies experiencing financial difficulties tend to adopt aggressive tax strategies in an effort to alleviate their tax burdens. Study Contributions: This research aims to contribute to discussing tax aggressiveness among Brazilian companies during economic crises, pointing out discrepancies with previous studies. Additionally, it seeks to broaden the understanding of tax aggressiveness in situations of crisis and financial difficulty, thereby enhancing the body of knowledge in the tax domain.

Time to transfer: Long-term effects of a sustained and spiraled content literacy intervention in the elementary grades.

We investigated the effectiveness of a sustained and spiraled content literacy intervention that emphasizes building domain and topic knowledge schemas and vocabulary for elementary-grade students. The model of reading engagement intervention underscores thematic lessons that provide an intellectual structure for helping students connect new learning to a general schema in Grade 1 (animal survival), Grade 2 (scientific investigation of past events like dinosaur mass extinctions), and Grade 3 (scientific investigation of living systems). A total of 30 elementary schools (N = 2,870 students) were randomized to a treatment or control condition. In the treatment condition (i.e., full spiral curriculum), students participated in content literacy lessons from Grades 1 to 3 during the school year and wide reading of thematically related informational texts in the summer following Grades 1 and 2. In the control condition (i.e., partial spiral curriculum), students participated in lessons in only Grade 3. The Grade 3 lessons for both conditions were implemented online during the COVID-19 pandemic school year. Results reveal that treatment students outperformed control students on science vocabulary knowledge across all three grades. Furthermore, intent-to-treat analyses revealed positive transfer effects on Grade 3 science reading (ES = .14), domain-general reading comprehension (ES = .11), and mathematics achievement (ES = .12). Treatment impacts were sustained at 14-month follow-up on Grade 4 reading comprehension (ES = .12) and mathematics achievement (ES = .16). Findings indicate that a content literacy intervention that spirals topics and vocabulary across grades can improve students' long-term academic achievement outcomes. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Learning Robust Shape Regularization for Generalizable Medical Image Segmentation.

Generalizable medical image segmentation enables models to generalize to unseen target domains under domain shift issues. Recent progress demonstrates that the shape of the segmentation objective, with its high consistency and robustness across domains, can serve as a reliable regularization to aid the model for better cross-domain performance, where existing methods typically seek a shared framework to render segmentation maps and shape prior concurrently. However, due to the inherent texture and style preference of modern deep neural networks, the edge or silhouette of the extracted shape will inevitably be undermined by those domain-specific texture and style interferences of medical images under domain shifts. To address this limitation, we devise a novel framework with a separation between the shape regularization and the segmentation map. Specifically, we first customize a novel whitening transform-based probabilistic shape regularization extractor namely WT-PSE to suppress undesirable domain-specific texture and style interferences, leading to more robust and high-quality shape representations. Second, we deliver a Wasserstein distance-guided knowledge distillation scheme to help the WT-PSE to achieve more flexible shape extraction during the inference phase. Finally, by incorporating domain knowledge of medical images, we propose a novel instance-domain whitening transform method to facilitate a more stable training process with improved performance. Experiments demonstrate the performance of our proposed method on both multi-domain and single-domain generalization.

DM-Fusion: Deep Model-Driven Network for Heterogeneous Image Fusion.

Heterogeneous image fusion (HIF) is an enhancement technique for highlighting the discriminative information and textural detail from heterogeneous source images. Although various deep neural network-based HIF methods have been proposed, the most widely used single data-driven manner of the convolutional neural network always fails to give a guaranteed theoretical architecture and optimal convergence for the HIF problem. In this article, a deep model-driven neural network is designed for this HIF problem, which adaptively integrates the merits of model-based techniques for interpretability and deep learning-based methods for generalizability. Unlike the general network architecture as a black box, the proposed objective function is tailored to several domain knowledge network modules to model the compact and explainable deep model-driven HIF network termed DM-fusion. The proposed deep model-driven neural network shows the feasibility and effectiveness of three parts, the specific HIF model, an iterative parameter learning scheme, and data-driven network architecture. Furthermore, the task-driven loss function strategy is proposed to achieve feature enhancement and preservation. Numerous experiments on four fusion tasks and downstream applications illustrate the advancement of DM-fusion compared with the state-of-the-art (SOTA) methods both in fusion quality and efficiency. The source code will be available soon.