Internal Knowledge Representation Research Articles

On the role of the UMLS in supporting diagnosis generation proposed by Large Language Models

Objective:Traditional knowledge-based and machine learning diagnostic decision support systems have benefited from integrating the medical domain knowledge encoded in the Unified Medical Language System (UMLS). The emergence of Large Language Models (LLMs) to supplant traditional systems poses questions of the quality and extent of the medical knowledge in the models’ internal knowledge representations and the need for external knowledge sources. The objective of this study is three-fold: to probe the diagnosis-related medical knowledge of popular LLMs, to examine the benefit of providing the UMLS knowledge to LLMs (grounding the diagnosis predictions), and to evaluate the correlations between human judgments and the UMLS-based metrics for generations by LLMs. Methods:We evaluated diagnoses generated by LLMs from consumer health questions and daily care notes in the electronic health records using the ConsumerQA and Problem Summarization datasets. Probing LLMs for the UMLS knowledge was performed by prompting the LLM to complete the diagnosis-related UMLS knowledge paths. Grounding the predictions was examined in an approach that integrated the UMLS graph paths and clinical notes in prompting the LLMs. The results were compared to prompting without the UMLS paths. The final experiments examined the alignment of different evaluation metrics, UMLS-based and non-UMLS, with human expert evaluation. Results:In probing the UMLS knowledge, GPT-3.5 significantly outperformed Llama2 and a simple baseline yielding an F1 score of 10.9% in completing one-hop UMLS paths for a given concept. Grounding diagnosis predictions with the UMLS paths improved the results for both models on both tasks, with the highest improvement (4%) in SapBERT score. There was a weak correlation between the widely used evaluation metrics (ROUGE and SapBERT) and human judgments. Conclusion:We found that while popular LLMs contain some medical knowledge in their internal representations, augmentation with the UMLS knowledge provides performance gains around diagnosis generation. The UMLS needs to be tailored for the task to improve the LLMs predictions. Finding evaluation metrics that are aligned with human judgments better than the traditional ROUGE and BERT-based scores remains an open research question.

UNDERSTANDING LARGE LANGUAGE MODELS: THE FUTURE OF ARTIFICIAL INTELLIGENCE

The article examines the newest direction in artificial intelligence - Large Language Models, which open a new era in natural language processing, providing the opportunity to create more flexible and adaptive systems. With their help, a high level of understanding of the context is achieved, which enriches the user experience and expands the fields of application of artificial intelligence. Large language models have enormous potential to redefine human interaction with technology and change the way we think about machine learning. An overview of the historical development of large language models is carried out, leading companies engaged in scientific research and development of effective systems are indicated. Information is provided regarding the internal structure and representation of knowledge in models. The main principles of learning are highlighted: data collection and their pre-processing, selection of an appropriate neural network architecture used in large language models. It is noted that the greatest progress has been achieved using the Transformer neural network, which is based on the mechanism of attention. The steps that significantly contribute to training, post-training, and optimizing the speed of training are highlighted. To evaluate the effectiveness and quality of language models, various metrics are used, which depend on the task to be solved. However, despite their advantages, large language models today are not without problems. The possibility of generating false information, fabricated facts, and unethical remarks presents a challenge for researchers and developers. It is important to continue work on increasing the responsibility of models, develop effective content filtering methods, and improve learning mechanisms. Understanding these problems and finding solutions to them are key steps towards building more efficient and reliable large language models. Openness, collective participation and dialogue between society, the scientific community and developers are becoming an integral part of ensuring the sustainable development of this technology.

KinRob: An ontology based robot for solving kinematic problems

Intelligent answering technology, which enables computers to solve problems automatically, is often used to develop tutorial systems, and has a wide range of application prospects. However, due to the lack of linguistic analysis and understanding methods, there are few researches on intelligent algorithms for solving kinematics problems. Developing such an algorithm is challenging, because solving kinematics problems is a complex task that includes text understanding, problem analysis, and automatic solution. To understand all these complexities involved in kinematics problems requires background knowledge. And only when an automatic solver contains a powerful internal knowledge representation system can it perform these tasks. We, thus, develop KinRob, an tutorial system for solving kinematics problems by combining neural network and ontology. Firstly, we propose an ontology for KinRob, which defines the knowledge of kinematics, and can help the robot understand a kinematics problem. Secondly, to match the text in natural language with the ontology, we propose a novel tagging scheme based on the kinematic problem understanding model in named entity recognition (NER). Finally, extensive experiments are conducted, and the experimental results show that the performance of the proposed method on a dataset of kinematic problems from authoritative sources better than the baseline algorithms.

Capsule-inferenced Object Detection for Remote Sensing Images

Frequent and accurate object detection based on remote sensing images can effectively monitor dynamic objects on the Earth's surface. While the Detection Transformer (DETR) offers a simple encoder-decoder structure and a direct set prediction approach to object detection, it falls short in complex remote sensing scenes where entity information and relative positions between objects are critical to target reasoning. Notably, the DETR model's feedforward neural network (FFN) relies on weighted summation for target reasoning, disregarding interactive feature information, which is a major factor affecting detection effectiveness. To address these shortcomings, we propose a DETR-based detection model called (CI_DETR), which uses capsule inference to improve remote sensing object detection. Our approach adds a Multi-level Feature Fusion module to the DETR network, allowing the network to learn how to spatially alter features at different levels, preserving only beneficial information for combination. In addition, we introduce a capsule reasoning module to mine entity information during inference and more effectively model the hierarchical correlation of internal knowledge representation in the neural network, consistent with the thinking model of the human brain. Lastly, we employ a sausage model to measure the similarities and differences of capsules, projecting them onto a curved surface for nonlinear function approximation and maximum preservation of the local responsiveness of capsule entities. Our experiments on public datasets confirm the superior detection performance of our proposed algorithm relative to many current detectors.

Architecture of network knowledge base of a complex military system

The article examines the architecture of a networked knowledge base and the organizational structure of a complex military-purpose system, which is built when a group of troops (forces) is created and kept in a state where it is capable of solving the tasks assigned to it. This requires a deep study of issues not only of modern tactics regarding the preparation and conduct of hostilities, but also more complex issues of scientific substantiation of the architecture of a networked knowledge base and the structure of a complex military system with a networked knowledge base. The internal representation of knowledge in the knowledge base (formal programmatic and logical content) is advisable to implement in the form of an adjacency matrix, which displays the relationship and relationship between target settings; initial conditions; the resources of the grouping of troops (temporary, material, combat and quantitative composition), their costs and replenishment; rules for the use of resources and the choice of criteria for their distribution. The knowledge base synthesizes a mathematical network model for making decisions, which provides a change (correction) of the structure of target attitudes when replenishing the knowledge base. Tasks solved in the knowledge base: selection of vertices and relations when replenishing catalogs; making changes to the adjacency matrix in accordance with the identified or changed relationships between targets. A necessary element of the synthesis of a mathematical network model for making decisions on the preparation and conduct of hostilities is the construction of the structure of the target systems of the system for a specific situation. A feature of controlling the correctness of knowledge presented in the form of target attitudes is the need for a joint analysis of the entire set of target attitudes and initial conditions in their relationship. For this, the matrix of the relations of target attitudes and the matrix of the relations of initial conditions are combined. The control of the correctness of the knowledge base is carried out when replenishing the knowledge base, it includes: identification of contradictions in the structure of target attitudes when making changes to this structure; search and detection of contradictions in the graph of the semantic network according to available resources and time; checking the completeness of the graph of the mathematical network model; issuance of revealed contradictions to an expert and their elimination. A practical approach to building the architecture of a networked knowledge base and the organizational structure of a complex military system can be implemented during the substantiation of the components and elements of the system when creating a grouping of troops (forces).

Towards a Context-Dependent Framework for Visualizing Mental Models

Mental models describe an internal representation of knowledge, which can be used to describe, explain, and predict a person’s interactions with their environment. While there is no shortage on mental model research, a recent literature review revealed the majority of research on mental models does not include an actual model of any kind. Mental models are able to capture complex information, which is difficult to effectively convey through written word alone. A set of visual metaphors are needed to effectively convey the disparate information that can be identified in mental models. We have conducted initial research to develop a notional framework for visualizing mental models based on the context of the elicitation methods used (and the data collected).

A Context-Driven Framework for Selecting Mental Model Elicitation Methods

Mental models describe an internal representation of knowledge of an individual or group, which can be used to interpret interactions with their environment and provide insight into decision-making strategies and prediction of performance. There are several ways to elicit mental models and analyze them; however, there is little guidance for selecting an appropriate elicitation method. Depending on different constraints of research and desired outcomes, different elicitation methods are more appropriate than others. Three criteria were identified as useful for selecting an appropriate elicitation method. These were the interaction level with participants, the number of participants being evaluated, and the resulting level of analytical detail that is required. A process for selecting the most appropriate mental model elicitation method is herein presented. Additionally, an overview of the factors that affect the selection of the mental models, and the different types of mental models are also presented.

МОДЕЛЬ ПРЕДСТАВЛЕННЯ ЕКСПЕРТНИХ ЗНАНЬ В ІНТЕЛЕКТУАЛЬНІЙ СИСТЕМІ ДІАГНОСТИКИ ЗАХВОРЮВАНЬ

In this article the model of representation of expert knowledge is offered in the intellectual system of diagnostics of diseases. The problem of representation of knowledge is considered. The use of rule-based model is reasonable. Determination over of linguistic variable is brought. Description of knowledge of the system is presented as linguistic variables. A fragment over of description of linguistic variables of the system is brought. The base of fuzzy rules of the system is described. Description of form of fuzzy rules is presented. Examples of external and internal representation of knowledge are made.

A Process Ontology for a Confectionery Service Robot

Confectionery items play a vital role in our lives. A plethora of confectionery items are ordered and consumed during various social gatherings, by a wide range of customers. The order may vary in terms of quantity, type of event and level of specificity. Moreover, some order may contain all the information needed while some may only contain the name of the festival for which suitable items need to be inferred and packaged accordingly. To understand all these intricacies for an automated agent involved in picking and packaging of items requires background knowledge. An automated system can perform these tasks only if, it comprises of a robust internal knowledge representation system. We, thus, propose an ontology for a service robot, SweetBot that defines pertinent concepts applicable to the confectionery domain needed by the robot in order to perform the picking and packaging of sweet items efficiently. We further evaluated the efficiency or the robustness of the model by using various SPARQL queries.

A rational cognitive architectural model of language generation

In this paper, a prototype model demonstrating the rationality in humans is developed. For queries to be answered, the internal knowledge representation mechanism is understood to be in the form of inter-related clusters permanently stored in the long term memory (LTM). In response to a query involving a single predicate (simplest case), only one cluster along with its various attributes and values may be retrieved. In contrast, a complex query involving multiple predicates requires the retrieval of multiple related chunks, forming a network of knowledge, often referred to as a concept map, from which the response is generated. An internal process that happens is the matching of predicate of the question with the facts in the knowledge-base and retrieval of the chunk(s) with the corresponding Attribute-Value pairs, through a process akin to unification, a substitution procedure. However, an unanswered query due to the absence of a specific knowledge generates a negative response.

Judgments of Learning for Words in Vertical Space.

Close relationship between physical space and internal knowledge representations has received ample support in the literature. For example, location of visually perceived information in vertical space has been shown to affect different numerical judgments. In addition, physical dimensions, such as weight or font size, were shown to affect judgments of learning (JOLs, an estimation of the likelihood that an item will be remembered later, or its perceived memorability). In two experiments we tested the hypothesis that differences in positioning words in vertical space may affect their perceived memorability, i.e., JOLs. In both Experiments, the words were presented in lower or in upper screen locations. In Experiment 1, JOLs were collected in the centre of the screen following word presentation. In Experiment 2, JOLs were collected at the point of word presentation and in the same location. In both experiments participants completed a free recall test. JOLs were compared between different vertically displaced presentation locations. In general, Bayesian analyses showed evidence in support for the null effect of vertical location on JOLs. We interpret our results as indicating that the effects of physical dimensions on JOLs are mediated by subjective importance, information that vertical location alone fails to convey.

Teacher learning communities: the landscape of subject leadership

Purpose – Teacher learning communities (TLCs) formed within subject departments are conceptualized as reform platform for facilitating school improvement and teacher development. The purpose of this paper is to examine the ways in which three TLCs were structured and managed for generating change capability in secondary schools in Hong Kong. Design/methodology/approach – A review of the literature on the contribution of learning communities to teacher professional development, and the functions and styles of subject leadership provides the theoretical background of the study. Participant observations and semi-structured interviews were employed as the main instruments for the investigation into the potentials of TLCs for enhancing professional renewals. Narrative accounts were obtained from a total of 21 teachers (including three subject leaders) on the nature and process of their participation in the renewal projects. Findings – The findings of the interview study reveal the landscape of subject leadership orientations that shaped the collaborative practices, power relations and knowledge representations within TLCs. Variability is noted in the subject leaders’ concerns for harmonious relationship, executive control, professional growth and leadership empowerment. The spectrum of knowledge structure in these communities is marked by a hierarchical relationship on one end, and a heterarchical relationship on the other. Research limitations/implications – This paper portrays the dynamics within three TLCs, and depicts a continuum of three leadership styles. The literature on subject leadership in TLCs could be enriched with further investigations into the contextual variables and sources of influence that affect the choice of leadership orientations of subject leaders who grapple with a balance between managerial imperative for control and accountability and a learning imperative for participative explorations that satisfy teachers’ aspirations for autonomy and creative investments in their own professional enhancement. Practical implications – Subject leadership training should emcompass reflective discourse on epistemological beliefs and assumptions on teacher learning and development, and leadership orientations that shape collaborative professional renewal practices. Originality/value – While research studies conducted in the Western world show that reflective and collaborative inquiries within professional communities have helped to improve classroom practices and teacher development (e.g. Hord, 1997), yet little is known about the management practices of subject leaders that shape the internal cultural conditions, power relations and knowledge representation within their departments in the context of Hong Kong.

The Effects of Idealized and Grounded Materials on Learning, Transfer, and Interest: An Organizing Framework for Categorizing External Knowledge Representations

Research in both cognitive and educational psychology has explored the effect of different types of external knowledge representations (e.g., manipulatives, graphical/pictorial representations, texts) on a variety of important outcome measures. We place this large and multifaceted research literature into an organizing framework, classifying three categories of external knowledge representations along a dimension of groundedness: (1) idealized, (2) grounded and including only relevant features, and (3) grounded and including irrelevant features. This organizing framework allows us to focus on the implications of these characteristics of external knowledge representations on three important educational outcomes: learning and immediate performance using the target knowledge, the degree to which that knowledge can transfer flexibly, and the interest engendered by the learning materials. We illustrate the framework by mapping a wide body of research from educational and cognitive psychology onto its dimensions. This framework can aid educators by clearly stating what the research literature says about these characteristics of external knowledge representations and how they activate and support the construction of internal knowledge representations. In particular, it will speak to how to best structure instruction using external knowledge representations with different characteristics, depending on the learning objective. Researchers will benefit from the analysis of the current state of knowledge and by the description of what open questions still remain.

Language learning in mindbodyworld: A sociocognitive approach to second language acquisition

Based on recent research in cognitive science, interaction, and second language acquisition (SLA), I describe a sociocognitive approach to SLA. This approach adopts anon-cognitivistview of cognition: Instead of an isolated computational process in which input is extracted from the environment and used to build elaborate internal knowledge representations, cognition is seen asadaptive intelligence,enabling our close and sensitivealignmentto our ecosocial environment in order to survive in it. Mind, body, and world are thus functionally integrated from a sociocognitive perspective instead of radically separated.Learning plays a major part in this scenario: If environments are ever-changing, then adaptation to them is continuous. Learning is part of our natural ability to so adapt, while retaining traces of that adaptation in the integrated mind-body-world system. Viewed in this way, SLA is adaptation to/engagement with L2 environments.Interactionalso plays a central role in sociocognitive SLA: We learn L2s through interacting with/in L2 environments. Founded on innate, universal skills which evolutionarilyprecededlanguage and make it possible, interaction supports SLA at every turn. Having presented this argument, I illustrate it by analyzing a video clip of an EFL tutoring session, indicating various ‘sociocognitive tools’ for interactive alignment which undergird L2 development.

Key Properties of Expert Movement Systems in Sport

This paper identifies key properties of expertise in sport predicated on the performer-environment relationship. Weaknesses of traditional approaches to expert performance, which uniquely focus on the performer and the environment separately, are highlighted by an ecological dynamics perspective. Key properties of expert movement systems include 'multi- and meta-stability', 'adaptive variability', 'redundancy', 'degeneracy' and the 'attunement to affordances'. Empirical research on these expert system properties indicates that skill acquisition does not emerge from the internal representation of declarative and procedural knowledge, or the imitation of expert behaviours to linearly reduce a perceived 'gap' separating movements of beginners and a putative expert model. Rather, expert performance corresponds with the ongoing co-adaptation of an individual's behaviours to dynamically changing, interacting constraints, individually perceived and encountered. The functional role of adaptive movement variability is essential to expert performance in many different sports (involving individuals and teams; ball games and outdoor activities; land and aquatic environments). These key properties signify that, in sport performance, although basic movement patterns need to be acquired by developing athletes, there exists no ideal movement template towards which all learners should aspire, since relatively unique functional movement solutions emerge from the interaction of key constraints.

Is the Mystery of Thought Demystified by Context‐Dependent Categorisation? Towards a New Relation Between Language and Thought

We argue that are no such things as literal categories in human cognition. Instead, we argue that there are merely temporary coalescences of dimensions of similarity, which are brought together by context in order to create the similarity structure in mental representations appropriate for the task at hand. Fodor (2000) contends that context‐sensitive cognition cannot be realised by current computational theories of mind. We address this challenge by describing a simple computational implementation that exhibits internal knowledge representations whose similarity structure alters fluidly depending on context. We explicate the processing properties that support this function and illustrate with two more complex models, one applied to the development of semantic knowledge (Rogers and McClelland, 2004), the second to the processing of simple metaphorical comparisons (Thomas and Mareschal, 2001). The models firstly demonstrate how phenomena that seem problematic for literal categorisation (such as the ‘non‐literal’ comparisons involved in metaphor and analogy) resolve to particular cases of the contextual modulation of mental representations; and secondly prompt a new perspective on the relation between language and thought: language affords the strategic control of context on semantic knowledge, allowing information to be brought to bear in a given situation that might otherwise not be available to influence processing. This may explain one way in which human thought is creative, and distinctive from animal cognition.

Measuring Internal Representations from Behavioral and Brain Data

The study of internal knowledge representations is a cornerstone of the research agenda in the interdisciplinary study of cognition. An influential proposal assumes that the brain uses its internal knowledge of the external world to constrain, in a top-down manner, high-dimensional sensory data into a lower-dimensional representation that enables perceptual decisions and other higher-level cognitive functions [1-9]. This proposal relies on a precise formulation of the observer-specific internal knowledge (i.e., the internal representations, or models) that guides reduction of the high-dimensional retinal input onto a low-dimensional code. Here, we directly revealed the content of subjective internal representations by instructing five observers to detect a face in the presence of only white noise, to force a pure top-down, knowledge-based task. We used reverse correlation methods to visualize each observer's internal representation that supports detection of an illusory face. Using reverse correlation again, this time applied to observers' electroencephalogram activity, we established where and when in the brain specific internal knowledge conceptually interprets the input white noise as a face. We show that internal representations can be reconstructed experimentally from behavioral and brain data, and that their content drives neural activity first over frontal and then over occipitotemporal cortex.

UMIS: A Service for User Model Interoperability

In this paper we describe UMIS, a service architecture that enables user adaptive applications to exchange User Model data on the Web. UMIS provides a set of facilities that allow applications to interoperate with minimum changes in their internal logics and knowledge representation. The goal is to support the process of interoperability in three ways: providing an efficient centralized discovery service; offering a service for simple interaction for the exchange of UM value in a p2p way; and offering a negotiation mechanism to be used in case of communication hurdles (i.e. semantic ambiguities and missing response). We developed a proof-of-concept prototype of UMIS and we tested it with an existing user-adaptive application. According to our test results, our approach improves the communication with respect to standard solutions for interoperability regarding the quality of exchange, with a negligible impact on the communication costs and traffic generation.

A Computational Simulation of the Cognitive Process of Children Knowledge Acquisition and Memory Development

The cognitive mechanisms of knowledge representation, memory establishment, and learning are fundamental issues in understanding the brain. A basic approach to studying these mental processes is to observe and simulate how knowledge is memorized by little children. This paper presents a simulation tool for knowledge acquisition and memory development for young children of two to five years old. The cognitive mechanisms of memory, the mathematical model of concepts and knowledge, and the fundamental elements of internal knowledge representation are explored. The cognitive processes of children’s memory and knowledge development are described based on concept algebra and the object-attribute-relation (OAR) model. The design of the simulation tool for children’s knowledge acquisition and memory development is presented with the graphical representor of memory and the dynamic concept network of knowledge. Applications of the simulation tool are described by case studies on children’s knowledge acquisition about family members, relatives, and transportation. This work is a part of the development of cognitive computers that mimic human knowledge processing and autonomous learning.

The Cognitive Process of Comprehension

Comprehension is an ability to understand the meaning of a concept or an action. Comprehension is an important intelligent power of abstract thought and reasoning of humans or intelligent systems. It is highly curious to explore the internal process of comprehension in the brain and to explain its basic mechanisms in cognitive informatics and computational intelligence. This paper presents a formal model of the cognitive process of comprehension. The mechanism and process of comprehension are systematically explained with its conceptual, mathematical, and process models based on the Layered Reference Model of the Brain (LRMB) and the Object-Attribute-Relation (OAR) model for internal knowledge representation. Contemporary denotational mathematics such as concept algebra and Real-Time Process Algebra (RTPA) are adopted in order to formally describe the comprehension process and its interaction with other cognitive processes of the brain.