Although most of the research on large language models (LLMs) focuses on their development and validation against datasets, significant gaps remain in their application to real-world knowledge-intensive tasks. This research addresses key challenges in using LLMs for extracting and synthesizing knowledge from unstructured sources, focusing on applications where the validity and consistency of the results are critical. We propose Ontology-based Adaptive Knowledge System (OAKS) as a holistic approach to manage the complexities of acquiring unstructured knowledge, varying user expertise, and dynamic query formulation. This research provides practical value for enabling the domain user community to leverage their technical documentation and expertise and accelerate ongoing working projects through improved literature review and cross-disciplinary insight discovery. Validated through empirical studies, our findings offer insight into best practices for the deployment of OAKS, bridging the gaps between AI capabilities and real-world needs in knowledge acquisition and research.

Recently, there has been a lot of excitement surrounding the use of reinforcement learning for robot control and navigation. However, many of these algorithms encounter difficulty navigating long or complex trajectories. This paper presents a new mobile robot control system called Stochastic Neural Control (SNC), that uses a stochastic policy gradient algorithm for local control and a modified probabilistic roadmap planner for global motion planning. In SNC, each mobile robot control decision is conditioned on observations from the robot sensors as well as pointcloud data, allowing the robot to safely operate within geometrically complex environments. SNC is tested on a number of challenging navigation tasks and learns advanced policies for navigation, collision-avoidance and fall-prevention. Three variants of the SNC system are evaluated against a conventional motion planning baseline. SNC outperforms the baseline and four other similar RL navigation systems in many of the trials. Finally, we present a strategy for transferring SNC from a simulated environment to a real robot. We empirically show that the SNC system exhibits good policies for mobile robot navigation when controlling a real mobile robot.

We present an end-to-end workflow to generate annotated image datasets for object detection semi-automatically, thereby reducing manual annotation need. With this workflow, which we call assistive learning, we are able to reduce manual annotation time on two experimental datasets by approximately 80%. The experimental results of this work show three contributions of the assistive learning workflow: (1) Savings on human annotation time; (2) generalizability to variable dataset sizes, domains and convolutional neural network (CNN) models; and (3) faster CNN training with limited amount of labeled data using a novel contextual sampling method, thereby a reduction in human workload early on in the assistive learning process. In addition, we wrap the workflow in an interactive annotation interface, allowing annotators without any machine learning experience to speed up the annotation process for training the CNN models.

The ability to access existing patents and patent-related information can have significant impacts on technology development and commercial markets. While such information is now available online, requesting the information about a specific technology remains a nontrivial task. The reason is that the online information often resides in different storage sources, is based on different modeling paradigms, and is managed by different independent entities. Ontology is one modeling paradigm that can capture semantically rich knowledge about a specific domain and its existing patents. That knowledge can be used to facilitate information sharing and information interoperability. This paper focuses on the bio-medical domain, where numerous patents have been filed in recent years. Specifically, this paper describes a computational framework that can facilitate the retrieval of intellectual property (IP-) related information in the bio-medical domain.

Automation in facility management and construction could significantly improve efficiency and productivity of the building industry. However, for robots and autonomous systems to operate effectively in dynamic and unstructured environments such as construction sites, they must be able to infer or obtain a semantic model of the environment. We propose a system for identifying common objects in a worksite and automatically adding them to an existing geometric model of the environment. The system is composed of two components: A novel 2D-3D object detection network designed to detect and localize a worksite objects, and a multi-object Kalman filter tracking system used to filter false-positive detections. The proposed system is validated using data collected with a purpose-built mobile robot. The mobile robot captures images of the worksite with an RGB-D camera and uses the proposed system to spawn newly placed objects in an existing geometric model of the worksite environment. The annotated RGB-D images are made publicly available to accelerate future research in this field.

IoT technology can have a huge impact in engineering by leveraging state-of-the-art information and communication technologies (ICT). In practice, however, it is challenging for IoT platforms to handle domain-specific engineering information (e.g., geometric model, engineering simulation model, etc.) along with sensor data of different types. Engineering information and sensor data need to be integrated for effective data utilization and shared and interoperable among a wide variety of software tools. This paper describes an IoT platform that is tailored to engineering applications and adopts an information modeling approach to facilitate data interoperability and to integrate engineering information with sensor data. In addition, a decentralized data management framework is employed so that the data owned by different project participants can be shared among authorized users and software agents. The IoT platform is demonstrated using a civil infrastructure monitoring scenario which involves various types of sensor data, as well as engineering models. The result shows that the IoT platform can facilitate information sharing and data utilization, in particular, for the civil infrastructure monitoring application.

Automatic localization of defects in metal castings is a challenging task, owing to the rare occurrence and variation in appearance of defects. Convolutional neural networks (CNN) have recently shown outstanding performance in both image classification and localization tasks. We examine how several different CNN architectures can be used to localize casting defects in X-ray images. We take advantage of transfer learning to allow state-of-the-art CNN localization models to be trained on a relatively small dataset. In an alternative approach, we train a defect classification model on a series of defect images and then use a sliding classifier method to develop a simple localization model. We compare the localization accuracy and computational performance of each technique. We show promising results for defect localization on the GRIMA database of X-ray images (GDXray) dataset and establish a benchmark for future studies on this dataset.

The use of data-driven predictive models is becoming increasingly popular in engineering and manufacturing sectors. This paper discusses the deployment of Gaussian Process Regression (GPR) predictive models for smart manufacturing. A scoring engine is developed based on the Predictive Model Markup Language (PMML) standard to illustrate the portability of predictive models among different statistical tools and different platforms. Specifically, we evaluate the tradeoffs between embedding GPR-based predictive models on a physical device and executing the predictive models on a managed cloud platform like the Google Compute Engine. We compare the performance of the two deployment strategies with two predictive models, namely an energy consumption model and a milling tool condition model, that are built with data from a Mori Seiki CNC milling machine. We describe how the response time of the two deployment strategies is related to the network latency and computational speed of the scoring machine hardware. It is shown that the time required to calculate model predictions is a significant factor in the overall response time of the embedded scoring engine. We demonstrate that the scoring engine on the cloud platform can achieve a lower response time and higher prediction rate than the microcomputer, due to the superior computational performance of the cloud-based hardware.

This paper describes a real-time data collection framework and an adaptive machining learning method for constructing a real-time energy prediction model for a machine tool. To effectively establish the energy consumption pattern of a machine tool over time, the energy prediction model is continuously updated with new measurement data to account for time-varying effects of the machine tool, such as tool wear and machine tool deterioration. In this work, a real-time data collection and processing framework is developed to retrieve raw data from a milling machine tool and its sensors and convert them into relevant input features. The extracted input features are then used to construct the energy prediction model using Gaussian Process (GP) regression. To update the GP regression model with real-time streaming data, we investigate the use of sparse representation of the covariance matrix to reduce the computational and storage demands of the GP regression. We compare computational efficiency of sparse GP to that of full GP regression model and show the effectiveness of the sparse GP regression model for tracking the variation in the energy consumption pattern of the target machine.

Recent advances in machine automation and sensing technology offer new opportunities for continuous condition monitoring of an operating machine. This paper describes an intelligent machine monitoring framework that integrates and utilizes data collection, management, and analytics to derive an adaptive predictive model for the energy usage of a milling machine. This model is designed using a Gaussian Process (GP) regression algorithm, which is a flexible regression method that also provides an uncertainty estimate. To improve computational efficiency, we propose a Collective Gaussian Process (CGP) in which the overall energy prediction is made by constructing local GP models weighted by probability distribution functions obtained using the Gaussian Mixture Model (GMM) technique. Finally, we demonstrate the ability of the proposed monitoring framework to construct an energy prediction model to predict the energy used to machine a part.

ABSTRACT Building occupants perceive information about the surrounding environment and navigate for safety during emergency evacuation (egress). They often choose their actions by interacting with their social groups and observing the actions of the surrounding crowd. Most egress simulators, however, ignore individuals' crowd perceptions and social interactions. This paper presents a novel platform, SAFEgress (Social Agent For Egress), in which building occupants are modeled as agents who decide their evacuation actions on the basis of their knowledge of the building and their interactions with the social groups and the neighboring crowd. Results from the SAFEgress prototype show that both agents' familiarity with the building and social influence can significantly impact egress performance. By simulating different occupants' behaviors, architects and facility managers may better understand the influence of human and social factors on evacuation and consequently design safer buildings and egress procedures. Copyright © 2014 John Wiley & Sons, Ltd.

In recent years, there has been an explosive growth in scientific and regulatory documents related to the patent system. Relevant information is siloed into many heterogeneous information domains making it very challenging to retrieve information across multiple domains. In this paper, we develop an ontology for the patent system to integrate information from the patent and court case domains. Through a use case erythropoietin, we demonstrate how this ontology can be used to enhance information retrieval across multiple domains.

For each industry, there exist many taxonomies that are intended for various applications. There are also multiple sources of regulations from different government agencies. Industry practitioners, unlike legal practitioners, are familiar with one or more industry-specific taxonomies but not necessarily regulatory organization systems. To help browsing of regulations by industry practitioners, we propose to map regulations to existing industry-specific taxonomies. A mapping from a single taxonomy to a single regulation is a trivial keyword matching task. From there, we examine techniques to map a single taxonomy to multiple regulations, as well as to map multiple taxonomies to a single regulation. Cosine similarity, Jaccard coefficient and market-basket analysis are tested to model the similarity metric between concepts from different taxonomies. Preliminary evaluations of the three metrics are performed. Examples from the building industry are drawn to illustrate the betterment of regulatory usage from the mapping between various taxonomies and regulations.

The complexity and diversity of government regulations make understanding the regulations a non-trivial task. One of the issues is the existence of multiple sources of regulations and interpretive guides; the latter are often independent of governing bodies. This work aims to develop an information infrastructure for legal information retrieval with applications to electronic-rulemaking. The pilot study focuses on accessibility regulations from the US Federal government, private organizations and European agencies. A shallow parser is developed to consolidate different regulations into a unified XML format, which is well suited for handling semi-structured data such as legal documents. Handcrafted rules and a text mining tool are developed to extract the important features, such as concepts, measurements, effective dates and so on, and to incorporate them into the corpus.To compare and locate related provisions from different regulatory documents, we employ Information Retrieval techniques to combine generic features with domain knowledge. Structural information from regulations, such as the hierarchical organization of provisions and heavy referencing among provisions, are used to help improve the relatedness analysis. Results are obtained to illustrate the use of regulatory structure and domain knowledge in provision comparisons. Application to an e-rulemaking scenario for a rights-of-way drafted regulation is shown to demonstrate extended capabilities of the prototype system.

The complexity and diversity of government regulations make understanding the regulations a non-trivial task. One of the issues is the existence of multiple sources of regulations and interpretive guides. In this work, we propose an information infrastructure for regulation analysis, which includes a document repository and tools for compliance assistance and similarity analysis. A regulatory repository is developed based on an XML format, and important features, such as concepts and measurements, are extracted using handcrafted rules and a text mining tool. Our framework provides compliance assistance using a reasoning tool based on First Order Predicate Calculus logic, where users are alerted of detected conflicts or otherwise compliance with the regulation. A relatedness analysis is performed by comparing the extracted features as well as structural and referential information from regulations. Examples of an electronic-rulemaking scenario and a compliance checking procedure are shown to demonstrate current capabilities of the prototype system.

This paper presents a loosely coupled service-composition paradigm. This paradigm employs a distributed data flow that differs markedly from centralized information flow adopted by current service integration frameworks, such as CORBA, J2EE and SOAP. Distributed data flows support direct data transmission to avoid many performance bottlenecks of centralized processing. In addition, active mediation is used in applications employing multiple web services that are not fully compatible in terms of data formats and contents. Active mediation increases the applicability of the services, reduces data communication among the services, and enables the application to control complex computations. The benefits of distributed data flow and active mediation are illustrated with various applications, such as dynamic type conversion, result extraction, and engineering application. It is shown that active mediation, combining with distributed data flows, can greatly improve the performance of an application utilizing multiple web services.

This paper focuses on the creation of a first order predicate calculus based regulation compliance-assistance system built upon an XML framework. Two areas of research that support the development of the compliance assistance system are discussed. The first is a document repository containing federal and state regulations and supplemental documents. The second is an XML framework for representing regulations and associated metadata. The prototype effort for the regulation assistance system has been focused on environmental regulations and related documents. The compliance assistance system is illustrated in the domain of used oil management. The objective of this research is to develop a formal infrastructure for regulatory information management and compliance assistance.

Government regulations are semi-structured text documents that are often voluminous, heavily cross-referenced between provisions and even ambiguous. Multiple sources of regulations lead to difficulties in both understanding and complying with all applicable codes. In this work, we propose a framework for regulation management and similarity analysis. An online repository for legal documents is created with the help of text mining tool, and users can access regulatory documents either through the natural hierarchy of provisions or from a taxonomy generated by knowledge engineers based on concepts. Our similarity analysis core identifies relevant provisions and brings them to the user's attention, and this is performed by utilizing both the hierarchical and referential structures of regulations to provide a better comparison between provisions. Preliminary results show that our system reveals hidden similarities that are not apparent between provisions based on node content comparisons.

Stable and robust finite element methods for the convective hydrodynamic model of semiconductor devices are developed and implemented on distributed memory parallel computers. Specifically, a stable and accurate space-time and Galerkin/least-squares finite element formulation is developed for the hydrodynamic transport equations for the conservation laws. In addition, Galerkin finite element methods are employed for the Poisson and lattice thermal diffusion equations. The inclusion of the lattice thermal diffusion equation in the numerical solution of the convective hydrodynamic model is presented for the first time. Numerical results for a bipolar transistor are included to illustrate the effectiveness of the numerical methods. Numerical simulations of semiconductor devices using the convective hydrodynamic model require a significant amount of computations. A single-program-multiple-data (SPMD) programming model is proposed for the implementation of the hydrodynamic model on distributed memory parallel computers. Employing the SPMD programming model, a serial program developed on a workstation can be converted into a parallel program with minimal changes. The parallel program has been ported to a wide range of distributed memory parallel computers including the iPSC/860 hypercube, the Touchstone Delta machine, and the IBM SP-1. Parallel performance results are reported for a bipolar transistor, silicon MESFET and diodes. The results indicate that the parallel hydrodynamic device simulator exhibits excellent speedups and scalability on distributed memory parallel computers with minimum communication overhead.

A symbolic node-addition model for matrix factorization of symmetric positive definite matrices is described. In this model, the nodes are added onto the filled graph one at a time. The advantage of the node-addition model is its simplicity and flexibility. The model can be immediately incorporated into finite element analysis programs. The model can also be extended to determine modification patterns in the matrix factors due to changes in the original matrix. For a given matrix K (= LDL t ), the time complexity of the algorithm for constructing the structure of the lower triangular matrix factor L is O (η( L )) where η( L ) is the number of nonzero entries in L .