Domain Knowledge Research Articles

Automated 3D convolutional neural network architecture design using genetic algorithm for pulmonary nodule classification

Cancer of the lungs is considered one of the primary causes of death among patients globally. Early detection contributes significantly to the success of pulmonary cancer treatment. To aid the pulmonary nodule classification, many models for the analysis of medical image utilizing deep learning have been developed. Convolutional neural network (CNN) recently, has attained remarkable results in various image classification tasks. Nevertheless, the CNNs performance is heavily dependent on their architectures which still heavily reliant on human domain knowledge. This study introduces a cutting-edge approach that leverages genetic algorithms (GAs) to automatically design 3D CNN architectures for differentiation between benign and malignant pulmonary nodules. The suggested algorithm utilizes the dataset of lung nodule analysis 2016 (LUNA16) for evaluation. Notably, our approach achieved exceptional model accuracy, with evaluations on the testing dataset yielding up to 95.977%. Furthermore, the algorithm exhibited high sensitivity, showcasing its robust performance in distinguishing between benign and malignant nodules. Our findings demonstrate the outstanding capabilities of the proposed algorithm and show an outstanding performance and attain a state of art solution in lung nodule classification.

Understanding the factors that influence CT utilization for mild traumatic brain injury in a low resource setting - a qualitative study using the Theoretical Domains Framework

IntroductionIn low resource settings (LRS), utilization of Computed Tomography scan (CTS) for mild traumatic brain injuries (mTBIs) presents unique challenges and considerations given the limited infrastructure, financial resources, and trained personnel. The Theoretical Domains Framework (TDF) offers a comprehensive theoretical lens to explore factors influencing the decision-making to order CTS for mTBI by imaging referrers (IRs). ObjectivesThe primary objective was to explore IRs’ beliefs about factors influencing CT utilization in mTBIs using TDF in Uganda.Differences in the factors influencing CTS ordering behavior across specialties, levels of experience, and hospital category were also explored. Materials and MethodsIn-depth semi-structured interviews guided by TDF were conducted among purposively selected IRs from 6 tertiary public and private hospitals with functional CTS services. A thematic analysis was performed with codes and emerging themes developed based on the TDF. ResultsEleven IRs including medical officers, non-neurosurgeon specialists and neurosurgeons aged on average 42 years (SD+/-12.3 years) participated.Identified factors within skills domain involved IRs’ clinical assessment and decision-making abilities, while beliefs about capabilities and consequences encompassed their confidence in diagnostic abilities and perceptions of CTS risks and benefits. The environmental context and resources domain addressed the availability of CT scanners and financial constraints. The knowledge domain elicited IRs’ understanding of clinical guidelines and evidence-based practices while social influences considered peer influence and institutional culture. For memory, attention & decision processes domain, IRs adherence to guidelines and intentions to order CT scans were cited. ConclusionUsing TDF, IRs identified several factors believed to influence decision making to order CTS in mTBI in a LRS. The findings can inform stakeholders to develop targeted strategies and evidence-based interventions to optimize CT utilization in mTBI such as; educational programs, workflow modifications, decision support tools, and infrastructure improvements, among others.

Classification of Dermoscopic Images Using CNN-SVM

Traditional machine learning methods like GLCM and ABCD rules have long been employed for image classification tasks. However, they come with inherent limitations, primarily the need for manual feature extraction. This manual feature extraction process is time-consuming and relies on expert domain knowledge, making it challenging for non-experts to use effectively. Deep learning methods, specifically Convolutional Neural Networks (CNN), have revolutionized image classification by automating the feature extraction. CNNs can learn hierarchical features directly from the raw pixel values, eliminating the need for manual feature engineering. Despite their powerful capabilities, CNNs have limitations, mainly when working with small image datasets. They may overfit the data or struggle to generalize effectively. In light of these considerations, this study adopts a hybrid approach that leverages the strengths of both deep learning and traditional machine learning. CNNs are automatic feature extractors, allowing the model to capture meaningful image patterns. These extracted features are then fed into a Support Vector Machine (SVM) classifier, known for its efficiency and effectiveness in handling small datasets. The results of this study are encouraging, with an accuracy of 0.94 and an AUC score of 0.94. Notably, these metrics outperform Abbas' previous research by a significant margin, underscoring the effectiveness of the hybrid CNN-SVM approach. This research reinforces that SVM classifiers are well-suited for tasks involving limited image data, yielding improved classification accuracy and highlighting the potential for broader applications in image analysis.

Breast Cancer Detection using CNN

Breast cancer stands as a leading cause of cancer- related fatalities worldwide. Assessing cancer accurately through eosin-stained images remains a complex task, often resulting in discrepancies among medical professionals while reaching a con- clusive diagnosis. To streamline this intricate process, Computer- Aided Diagnosis (CAD) systems present a promising avenue, aiming to reduce costs and enhance efficiency. Traditional clas- sification methods hinge on problem-specific feature extraction, rooted in domain knowledge. However, addressing the multitude of challenges posed by these feature-centric techniques has led to the emergence of deep learning methods as viable alternatives. Here, we propose a novel approach employing Convolutional Neural Networks (CNNs) for the classification of hematoxylin and eosin-stained breast biopsy images. Our method categorizes images into four distinct groups: normal tissue, benign lesion, in situ carcinoma, and invasive carcinoma. Additionally, it per- forms a binary classification distinguishing carcinoma from non- carcinoma cases. The meticulously designed network architecture facilitates information extraction across multiple scales, encom- passing both individual nuclei and overall tissue organization. This design choice enables seamless integration of our proposed system with wholeslide histology images. Notably, our method achieves a commendable accuracy of 77.8four-class classification and demonstrates a high sensitivity of 95.6% in identifying cancer cases.

Semantically Enriched Keyword Prefetching Based on Usage and Domain Knowledge

In intelligent web systems [2], web prefetching [27] plays a crucial role. In order to make accurate predictions for web prefetching, it is important but challenging to uncover valuable information from web use statistics [16]. Using statistics and domain expertise, this study presents a new approach dubbed SPUDK for efficient prefetching. In this paper, it is shown how web access logs can be used efficiently for browsing prediction. Our main focus is on the technique needed to manage the queries found in web access logs so that valuable information can be attained. We further process these access logs using a taxonomy and a thesaurus, WordNet, to find the semantics of queries. SPUDK, a system that organises use data into semantic clusters, is one example of this approach. Our contributions in this paper are as follows: (1) A technique to exploit query keywords from access logs. (2) An approach to enrich queries with semantic information. (3) A new similarity measure for finding similarity among URLs present in access logs. (4) A novel clustering technique to find semantic clusters of URLs. (5) Experimental evaluation of the proposed system. The proposed SPUDK system is evaluated using American Online (AOL) logs, which gives improvement of 39% in precision of prediction, 35% in hit ratio and reduction of 50.6% in latency on average as compared to other prediction techniques in the literature.

Disclaiming knowledge to encourage participation in research group meetings

This study investigates instances where experienced researchers make explicit claims of lack of knowledge (e.g., ‘I have to reveal my ignorance completely’) in the context of paper discussions during research group meetings. Drawing on multimodal interaction analysis, the analysis focuses on epistemic disclaimers in their sequential contexts, and the local management of institutional identities and domains of knowledge. The analyzed data draw from video-ethnographic work involving participant observation and video recordings at a research program in Physical Chemistry at a Swedish university. Focusing on epistemic disclaimers occurring in first pair-parts, the analysis explores how senior researchers employ disclaimers to mark the eliciting function of their questions, as they work to encourage participation by opening the floor while positioning co-participants as (more) knowledgeable. The results evidence how the accomplishment of peer collaboration and knowledge distribution in scientific work involves the management of rights and responsibilities with respect to knowledge and scientific expertise.

X-LoRA: Mixture of low-rank adapter experts, a flexible framework for large language models with applications in protein mechanics and molecular design

We report a mixture of expert strategy to create fine-tuned large language models using a deep layer-wise token-level approach based on low-rank adaptation (LoRA). Starting with a set of pre-trained LoRA adapters, our gating strategy uses the hidden states to dynamically mix adapted layers, allowing the resulting X-LoRA model to draw upon different capabilities and create never-before-used deep layer-wise combinations to solve tasks. The design is inspired by the biological principles of universality and diversity, where neural network building blocks are reused in different hierarchical manifestations. Hence, the X-LoRA model can be easily implemented for any existing large language model without a need for modifications of the underlying structure. We develop a tailored X-LoRA model that offers scientific capabilities, including forward/inverse analysis tasks and enhanced reasoning capability, focused on biomaterial analysis, protein mechanics, and design. The impact of this work includes access to readily expandable and adaptable models with strong domain knowledge and the capability to integrate across areas of knowledge. Featuring experts in biology, mathematics, reasoning, bio-inspired materials, mechanics and materials, chemistry, protein biophysics, mechanics, and quantum-mechanics based molecular properties, we conduct a series of physics-focused case studies. We examine knowledge recall, protein mechanics forward/inverse tasks, protein design, adversarial agentic modeling including ontological knowledge graph construction, and molecular design. The model is capable not only of making quantitative predictions of nanomechanical properties of proteins or quantum mechanical molecular properties but also reasoning over the results and correctly predicting likely mechanisms that explain distinct molecular behaviors.

An automatic completion method for design domain knowledge graph using surrogate model, for rapid performance evaluation

The data and information generated during the design process, such as part parameters, affect the efficiency and quality of product design. Knowledge graph (KG) is often used to express and reuse the above knowledge. However, due to the decentralised and complex nature of the KG, it lacks the quantitative knowledge required for part selection, e.g. the relationship between part type and performance. To complete the knowledge and meet the requirements for rapid and accurate performance evaluation, this paper proposes an automatic KGC method based on a surrogate model. Firstly, the schema layer is established based on the design knowledge ontology. For the sampling process, the query statement is applied to extract knowledge such as design parameters, which define the range of sample points. Secondly, the design of experiments (DOE) method can obtain the data set of the target performance through simulation. Surrogate models are built based on multiple machine learning algorithms. Meanwhile, the hyperparameter optimisation method can further improve the prediction accuracy of the model. Finally, the performance of the design scheme is predicted, and the results are used to complete the KG through knowledge extraction. The method is validated by the selection and evaluation of bogie parts.

Working together with people with intellectual disability to make a difference: a protocol for a mixed-method co-production study to address inequities in cervical screening participation

IntroductionCervical cancer is one of the most preventable cancers yet remains a disease of inequity for people with intellectual disability, in part due to low screening rates. The ScreenEQUAL project will use an integrated knowledge translation (iKT) model to co-produce and evaluate accessible cervical screening resources with and for this group.MethodsStage 1 will qualitatively explore facilitators and barriers to screening participation for people with intellectual disability, families and support people, healthcare providers and disability sector stakeholders (n ≈ 20 in each group). An accessible multimodal screening resource, accompanying supporting materials for families and support people, and trauma-informed healthcare provider training materials will then be co-produced through a series of workshops. Stage 2 will recruit people with intellectual disability aged 25 to 74 who are due or overdue for screening into a single-arm trial (n = 48). Trained support people will provide them with the co-produced resource in accessible workshops (intervention) and support them in completing pre-post questions to assess informed decision-making. A subset will participate in qualitative post-intervention interviews including optional body-mapping (n ≈ 20). Screening uptake in the 9-months following the intervention will be measured through data linkage. Family members and support people (n = 48) and healthcare providers (n = 433) will be recruited into single-arm sub-studies. Over a 4-month period they will, respectively, receive the accompanying supporting materials, and the trauma-informed training materials. Both groups will complete pre-post online surveys. A subset of each group (n ≈ 20) will be invited to participate in post-intervention semi-structured interviews.Outcomes and analysisOur primary outcome is a change in informed decision-making by people with intellectual disability across the domains of knowledge, attitudes, and screening intention. Secondary outcomes include: (i) uptake of screening in the 9-months following the intervention workshops, (ii) changes in health literacy, attitudes and self-efficacy of family members and support people, and (iii) changes in knowledge, attitudes, self-efficacy and preparedness of screening providers. Each participant group will evaluate acceptability, feasibility and usability of the resources.DiscussionIf found to be effective and acceptable, the co-produced cervical screening resources and training materials will be made freely available through the ScreenEQUAL website to support national, and potentially international, scale-up.

Self-attentive Rationalization for Interpretable Graph Contrastive Learning

Graph augmentation is the key component to reveal instance-discriminative features of a graph as its rationale – an interpretation for it – in graph contrastive learning (GCL). And existing rationale-aware augmentation mechanisms in GCL frameworks roughly fall into two categories and suffer from inherent limitations: (1) non-heuristic methods with the guidance of domain knowledge to preserve salient features, which require expensive expertise and lack generality, or (2) heuristic augmentations with a co-trained auxiliary model to identify crucial substructures, which face not only the dilemma between system complexity and transformation diversity, but also the instability stemming from the co-training of two separated submodels. Inspired by recent studies on transformers, we propose S elf-attentive R ationale guided G raph C ontrastive L earning (SR-GCL), which integrates rationale generator and encoder together, leverages the self-attention values in transformer module as a natural guidance to delineate semantically informative substructures from both node- and edge-wise perspectives, and contrasts on rationale-aware augmented pairs. On real-world biochemistry datasets, visualization results verify the effectivenes and interpretability of self-attentive rationalization, and the performance on downstream tasks demonstrates the state-of-theart performance of SR-GCL for graph model pre-training. Codes are available at https://github.com/lsh0520/SR-GCL .

Advancing language models through domain knowledge integration: a comprehensive approach to training, evaluation, and optimization of social scientific neural word embeddings

Advancing language models through domain knowledge integration: a comprehensive approach to training, evaluation, and optimization of social scientific neural word embeddings

Онтологизация познания: уровни, направления и средства реализации

The article investigates the problem of ontologization of cognition, i.e. increasing the level of conformity between the knowledge system and being. Various levels of ontologization of knowledge are examined depending on the nature of the reflection of reality in the system of knowledge. Four levels of ontology are distinguished: the first level corresponds to the injective and bijective mappings of reality onto the knowledge system; the second level involves surjective mapping; the third level encompasses knowledge formed through isomorphic mapping of relationships between elements from reality onto the knowledge system; the fourth level involves knowledge formed through isomorphic mapping of relationships between elements within the knowledge system. Prospects for the ontologization of cognition are related to the expansion of metaphysical and consequent knowledge, the elevation to the fourth level of knowledge domains that lack ontologicality, and the expansion of knowledge linking secondary laws and properties of objects to patterns. The main tool for ontologizing cognition should be a general theory of systems, extended with a metaphysical component and supplemented with extensive collections of patterns, primitives, secondary properties and laws. The author is working on such a general systems theory, referred to as the “empirico-metaphysical general theory of systems”. In conclusion, development of an empirico-metaphysical theory of systems promises to provide a robust framework for addressing the challenges of cognition representation, synthesis, and application across diverse domains. As researchers continue to delve into the nuances of ontological alignment, the proposed framework offers a promising avenue for fostering interdisciplinary collaboration and advancing our understanding of the fundamental nature of cognition.

Educational Transformation: Refocusing Research, Innovation, Technology Integration and Social Inclusion – An Editorial

In the post-pandemic era, the world simultaneously witnessed a rapid move towards technological integration and global interconnectedness, alongside social-economic disparities and setbacks. As a result of the COVID-19 pandemic, scientific progress extended beyond the healthcare industry and encompassed various domains of knowledge, including education. The pandemic accelerated the incorporation of ICT in education, an area that had lagged behind other sectors. Conversely, digital integration generates a disparity in knowledge acquisition between individuals who possess access to it and those who do not. This experience serves as a reminder that a considerable number of communities globally remain behind in terms of socioeconomic development and access to digital resources. This trend is especially evident in the global south, where the integration of numerous global educational innovations and approaches may further isolate an already marginalized community by failing to account for contextual factors. Providing this complexity of introducing innovative efforts to education, it is imperative to refocus contextual educational paradigms, employing a multimodal approach that intertwines research, innovation, technology integration, and social inclusion. The purpose of this editorial is to emphasize the importance of an all-encompassing educational approach that not only equips students for the challenges of the future but also cultivates a just, inventive, and resilient society. A total of 11 papers were selected to be published in this issue, which examines four significant domains of educational transformation: the integration of technology, innovative approaches, research in education, and the representation of perspectives from underrepresented groups in education. The increasing integration of technology in the field of education is evident in the articles that discuss ICT, media usage, and visual materials. Two papers revisit inquiry-based science instruction and school-home partnerships as educational strategies. Two publications centered on teachers' practice and intention research and action research exemplify the recent shift towards involving teachers in research activities, both in higher education and at the classroom level. Three additional papers investigate the perspectives of students and underprivileged groups who are often overlooked in the development of education procedures and policies. This journal issue, thus, represents the current focus, endeavors, and implementation of education across several educational disciplines. Teacher’s World: Journal of Education and Research, 49 (2),9-14

Combining ChatGPT and knowledge graph for explainable machine learning-driven design: a case study

Machine learning has been widely used in design activities, enabling more informed decision-making. However, high-performance machine learning models, often referred to as ‘black-box', result in a lack of explainability regarding predictions. The absence of explainability erodes the trust between designers and these models and hinders human-machine collaboration for desirable design decisions. Explainable AI focuses on creating explanations that are accessible and comprehensible to stakeholders, thereby improving explainability. A recent advancement in the field of explainable AI involves leveraging domain-specific knowledge via knowledge graph. Additionally, the advent of large language models like ChatGPT, acclaimed for their ability to output domain knowledge, perform complex language processing, and support seamless end-user interaction, has the potential to expand the horizons of explainable AI. Inspired by these developments, we propose the novel hybrid method that synergizes ChatGPT and knowledge graph to augment post-hoc explainability in design context. The outcome is the generation of more contextual and meaningful explanations, with the added possibility of further interaction to uncover deeper insights. The effectiveness of the proposed method is illustrated through a case study on customer segmentation.

Biologically informed NeuralODEs for genome-wide regulatory dynamics

BackgroundGene regulatory network (GRN) models that are formulated as ordinary differential equations (ODEs) can accurately explain temporal gene expression patterns and promise to yield new insights into important cellular processes, disease progression, and intervention design. Learning such gene regulatory ODEs is challenging, since we want to predict the evolution of gene expression in a way that accurately encodes the underlying GRN governing the dynamics and the nonlinear functional relationships between genes. Most widely used ODE estimation methods either impose too many parametric restrictions or are not guided by meaningful biological insights, both of which impede either scalability, explainability, or both.ResultsWe developed PHOENIX, a modeling framework based on neural ordinary differential equations (NeuralODEs) and Hill-Langmuir kinetics, that overcomes limitations of other methods by flexibly incorporating prior domain knowledge and biological constraints to promote sparse, biologically interpretable representations of GRN ODEs. We tested the accuracy of PHOENIX in a series of in silico experiments, benchmarking it against several currently used tools. We demonstrated PHOENIX’s flexibility by modeling regulation of oscillating expression profiles obtained from synchronized yeast cells. We also assessed the scalability of PHOENIX by modeling genome-scale GRNs for breast cancer samples ordered in pseudotime and for B cells treated with Rituximab.ConclusionsPHOENIX uses a combination of user-defined prior knowledge and functional forms from systems biology to encode biological “first principles” as soft constraints on the GRN allowing us to predict subsequent gene expression patterns in a biologically explainable manner.

OptiComm-GPT: a GPT-based versatile research assistant for optical fiber communication systems

With the increasing capacity and complexity of optical fiber communication systems, both academic and industrial requirements for the essential tasks of transmission systems simulation, digital signal processing (DSP) algorithms verification, system performance evaluation, and quality of transmission (QoT) optimization are becoming significantly important. However, due to the intricate and nonlinear nature of optical fiber communication systems, these tasks are generally implemented in a divide-and-conquer manner, which necessitates a profound level of expertise and proficiency in software programming from researchers or engineers. To lower this threshold and facilitate professional research easy-to-start, a GPT-based versatile research assistant named OptiComm-GPT is proposed for optical fiber communication systems, which flexibly and automatically performs system simulation, DSP algorithms verification, performance evaluation, and QoT optimization with only natural language. To enhance OptiComm-GPT’s abilities for complex tasks in optical fiber communications and improve the accuracy of generated results, a domain information base containing rich domain knowledge, tools, and data as well as the comprehensive prompt engineering with well-crafted prompt elements, techniques, and examples is established and performs under a LangChain-based framework. The performance of OptiComm-GPT is evaluated in multiple simulation, verification, evaluation, and optimization tasks, and the generated results show that OptiComm-GPT can effectively comprehend the user’s intent, accurately extract system parameters from the user’s request, and intelligently invoke domain resources to solve these complex tasks simultaneously. Moreover, the statistical results, typical errors, and running time of OptiComm-GPT are also investigated to illustrate its practical reliability, potential limitations, and further improvements.

Advancing Ontology Alignment in the Labor Market: Combining Large Language Models with Domain Knowledge

One of the approaches to help the demand and supply problem in the labor market domain is to change from degree-based hiring to skill-based hiring. The link between occupations, degrees and skills is captured in domain ontologies such as ESCO in Europe and O*NET in the US. Several countries are also building or extending these ontologies. The alignment of the ontologies is important, as it should be clear how they all relate. Aligning two ontologies by creating a mapping between them is a tedious task to do manually, and with the rise of generative large language models like GPT-4, we explore how language models and domain knowledge can be combined in the matching of the instances in the ontologies and in finding the specific relation between the instances (mapping refinement). We specifically focus on the process of updating a mapping, but the methods could also be used to create a first-time mapping. We compare the performance of several state-of-the-art methods such as GPT-4 and fine-tuned BERT models on the mapping between ESCO and O*NET and ESCO and CompetentNL (the Dutch variant) for both ontology matching and mapping refinement. Our findings indicate that: 1) Match-BERT-GPT, an integration of BERT and GPT, performs best in ontology matching, while 2) TaSeR outperforms GPT-4, albeit marginally, in the task of mapping refinement. These results show that domain knowledge is still important in ontology alignment, especially in the updating of a mapping in our use cases in the labor domain.

NREM3 Sleep Stage Estimation Based on Accelerometer by Body Movement Count and Biological Rhythms

This paper proposes the method by physiological knowledge to improve the estimation performance of the NREM3 sleep based on the waist-attached accelerometer. Specifically, this paper proposes the hybrid method that combines the method based on body movement counts and the method based on biological rhythms of sleep. Through the human subject experiment, the following implications were revealed: (1) the proposed method can outperform famous machine learning models (Random Forest and LSTM) trained with automatically generated features that do not sufficiently incorporate domain knowledge; (2) when the input features are based on domain knowledge, the estimator explicitly designed by humans can outperform the machine learning method; and (3) combining the body movement counting method and the biological rhythm-based method can suppress the error of the body movement counting method and reduce false positives.

Domain-specific Embeddings for Question-Answering Systems: FAQs for Health Coaching

FAQs are widely used to respond to users’ knowledge needs within knowledge domains. While LLM might be a promising way to address user questions, they are still prone to hallucinations i.e., inaccurate or wrong responses, which, can, inter alia, lead to massive problems, including, but not limited to, ethical issues. As a part of the healthcare coach chatbot for young Nigerian HIV clients, the need to meet their information needs through FAQs is one of the main coaching requirements. In this paper, we explore if domain knowledge in HIV FAQs can be represented as text embeddings to retrieve similar questions matching user queries, thus improving the understanding of the chatbot and the satisfaction of the users. Specifically, we describe our approach to developing an FAQ chatbot for the domain of HIV. We used a predefined FAQ question-answer knowledge base in English and Pidgin co-created by HIV clients and experts from Nigeria and Switzerland. The results of the post-engagement survey show that the chatbot mostly understood the user’s questions and could identify relevant matching questions and retrieve an appropriate response.

A Framework for Enhancing Behavioral Science Research with Human-Guided Language Models

Many behavioral science studies result in large amounts of unstructured data sets that are costly to code and analyze, requiring multiple reviewers to agree on systematically chosen concepts and themes to categorize responses. Large language models (LLMs) have potential to support this work, demonstrating capabilities for categorizing, summarizing, and otherwise organizing unstructured data. In this paper, we consider that although LLMs have the potential to save time and resources performing coding on qualitative data, the implications for behavioral science research are not yet well understood. Model bias and inaccuracies, reliability, and lack of domain knowledge all necessitate continued human guidance. New methods and interfaces must be developed to enable behavioral science researchers to efficiently and systematically categorize unstructured data together with LLMs. We propose a framework for incorporating human feedback into an annotation workflow, leveraging interactive machine learning to provide oversight while improving a language model's predictions over time.