Multi-level Representation Research Articles

HIERARCHICAL REPRESENTATION OF CAUSAL RELATIONSHIPS TO DETAIL EXPLANATIONS IN INTELLIGENT SYSTEMS

The subject of research in the article is the processes of constructing explanations in intelligent systems based on the use of causal dependencies. The aim is to develop a hierarchical representation of causal relationships between the actions of an intelligent system to form an explanation of the process of the system's operation with a given degree of generalization or detailing. Representation of the hierarchy of cause-and-effect relationships allows you to form an explanation at a given level of detail using the input data in the form of a temporally ordered sequence of events reflecting the known actions of an intelligent system. Tasks: structuring the hierarchy of cause-and-effect relationships for known variants of the decision-making process in an intelligent information system, considering the temporal ordering of the corresponding actions; development of a model of a multi-level representation of causal dependencies for description for explanations in an intelligent system. The approaches used are: counterfactual analysis of causality, used to describe alternative dependencies for possible decision-making options; linear temporal logic to reflect the temporal aspect of causation. The following results were obtained. A generalized hierarchy of cause-and-effect relationships is highlighted for the known variants of the process of obtaining recommendations in an intelligent information system based on the temporal ordering of the corresponding decision-making actions. A model of hierarchical representation of causal dependencies has been developed to describe explanations in an intellectual system with a given degree of detail. Conclusions. The scientific novelty of the results obtained is as follows. A model of hierarchical representation of time-ordered causal relationships is proposed to describe the explanations of the operation of an intelligent system with a given degree of detail. At the top level of the hierarchy, the model defines a generalized causal relationship between the event of using the input data and the event of the result of the system's operation. This connection describes the current task that the intelligent information system solves. At the lower level, cause-and-effect relationships are set between events sequential in time, between which there are no other events. At intermediate levels of the hierarchical representation, the causal dependencies of pairs of events are determined, between which there are other events. The developed model creates conditions for constructing explanations with a given degree of detailing of the actions of the decision-making process in an intelligent system. The model also provides the ability to describe early and late anticipation of alternative sequences of the decision-making process by describing causal dependencies for events between which there are other events.

A CNN-transformer hybrid approach for decoding visual neural activity into text

Background and ObjectiveMost studies used neural activities evoked by linguistic stimuli such as phrases or sentences to decode the language structure. However, compared to linguistic stimuli, it is more common for the human brain to perceive the outside world through non-linguistic stimuli such as natural images, so only relying on linguistic stimuli cannot fully understand the information perceived by the human brain. To address this, an end-to-end mapping model between visual neural activities evoked by non-linguistic stimuli and visual contents is demanded. MethodsInspired by the success of the Transformer network in neural machine translation and the convolutional neural network (CNN) in computer vision, here a CNN-Transformer hybrid language decoding model is constructed in an end-to-end fashion to decode functional magnetic resonance imaging (fMRI) signals evoked by natural images into descriptive texts about the visual stimuli. Specifically, this model first encodes a semantic sequence extracted by a two-layer 1D CNN from the multi-time visual neural activity into a multi-level abstract representation, then decodes this representation, step by step, into an English sentence. ResultsExperimental results show that the decoded texts are semantically consistent with the corresponding ground truth annotations. Additionally, by varying the encoding and decoding layers and modifying the original positional encoding of the Transformer, we found that a specific architecture of the Transformer is required in this work. ConclusionsThe study results indicate that the proposed model can decode the visual neural activities evoked by natural images into descriptive text about the visual stimuli in the form of sentences. Hence, it may be considered as a potential computer-aided tool for neuroscientists to understand the neural mechanism of visual information processing in the human brain in the future.

Text Sentiment Analysis of German Multilevel Features Based on Self-Attention Mechanism

In this paper, we propose a multilevel feature representation method that combines word-level features, such as German morphology and slang, and sentence-level features, such as special symbols and English-translated sentiment information, and build a deep learning model for German sentiment classification based on the self-attentive mechanism, in order to address the characteristics of German social media texts that are colloquial, irregular, and diverse. Compared with the existing studies, this model not only has the most obvious improvement effect but also has better feature extraction and classification ability for German emotion.

Developing Practical Investment Resilience

While seeking practical tools for resilient investors, a myopic ability for single-period decisions to reflect outcomes at multiple time horizons without dynamic programming is developed. It captures tail risk and, with multiple overlapping observations compounded at different frequencies, captures time-series properties as well. These benefits rely on pairing multilevel representation of the utility probability distribution with the use of Rubinstein utility—logarithmic in investor surplus. Other tools demonstrated for practical use include top-down hierarchical cluster analysis to provide qualitative information, simplified shrinkage of observed return means (but not covariances or higher moments), and better measures of potential tail risk. The resulting asset allocation examples drawn from 28 mutual funds and exchange-traded funds vary by investor risk aversion, degree of anticipation of different regimes, and whether they consider short, longer, or multiple time horizons. For single-horizon problems, resilient methods appear most helpful to conservative investors or those who trade infrequently. In contrast, multiple time horizons appear potentially helpful to all investors. <b>Key Findings</b> ▪ A toolset for investors seeking greater resilience against systemic risk stemming from nontraditional sources, such as pandemics, revolutionary technology, climate change, and political upheavals, is created, with an example based on the impact of poorly understood derivatives in the financial crisis of 2007–2009. ▪ Confronting the wavelike character of systemic disruptions with time stages that may include a sudden shock, rolling disruption, or gradual recovery as winners and losers are revealed leads to a method of making single-period allocations that reflect multiple time horizon outcomes while staying within a simple and easily implemented decision-theory framework. ▪ Other tools for resilient investing include top-down hierarchical cluster analysis, measures of tail risk customized by risk aversion, preserving more useful information by shrinking means but not covariances, and transparent open-source software.

Simulation of the business processes of the developing enterprise to create complex products with multi-level component architecture

The problem of research of business processes related to the organization and management of projects to create the new complex technical products in aerospace, shipbuilding, and other fields of mechanical engineering is stated and solved. The relevance of the study is related to the planning of a set of business processes, taking into account the architectural representation of the complex product being created. The study aims to develop a new method of synthesis of the multilevel complex of business processes to manage the creation of new technology, taking into account the multi-component representation of the complex product architecture. Much attention is paid to the complex product architecture, which is presented as a set of interconnected components. The following component types are outlined: components that can be reused; components that can be adapted to the requirements of the new product; innovative components. The types of components are studied in terms of the building of the set of business processes on their creation, taking into account the main indicators (costs, execution time, risks). Using the positive experience of past developments in the form of ready-to-use business processes helps to minimize the risks, and the new business solutions in their turn ensure the competitiveness of the complex product. The information base of positive past business processes in the form of precedents base is being built. The precedents base is used to find the close analogs of ready-to-use business processes. The set of business processes related to the management of complex product creation is formed. It has a multilevel representation that fully corresponds to the tree-like component architecture of the complex product. The method to synthesize the complexity of business processes that are planned to perform the tasks on the management of projects on creation of complex products for the developing enterprise is constructed. It is based on the “from top to bottom” system design principle. The optimization model to find the minimum costs, time, and risks in the process of business processes creation that takes into account the multilevel structure of the business process is built. The simulation event-based model in the form of a related agent set is used to study the successful execution of planned business processes related to the creation of the complex product. The results of this study should be used in the practice of creating innovative technical products that have a complex multi-level component architecture for the organization and planning of business processes related to development management. Mathematical methods used: systems analysis, precedent theory, lexicographic ordering, component design, integer optimization, simulation modeling.

Integrating artificial intelligence and analytics in smart grids: a systematic literature review

PurposeThe purpose of this study is to explore the latest approaches in integrating artificial intelligence and analytics (AIA) in energy smart grid projects. Empirical results are synthesized to highlight their relevance from a technology and project management standpoint, identifying several lessons learned that can be used for planning highly integrated and automated smart grid projects.Design/methodology/approachA systematic literature review leads to selecting 108 research articles dealing with smart grids and AIA applications. Keywords are based on the following research questions: What is the growth trend in Smart Grid projects using intelligent systems and data analytics? What business value is offered when AI-based methods are applied? How do applications of intelligent systems combine with data analytics? What lessons can be learned for Smart Grid and AIA projects?FindingsThe 108 selected articles are classified according to the following four research issues in smart grids project management: AIA integrated applications; AI-focused technologies; analytics-focused technologies; architecture and design methods. A broad set of smart grid functionality is reviewed, seeking to find commonality among several applications, including as follows: dynamic energy management; automation of extract, transform and load for Supervisory Control And Data Acquisition (SCADA) systems data; multi-level representations of data; the relationship between the standard three-phase transforms and modern data analytics; real-time or short-time voltage stability assessment; smart city architecture; home energy management system; building energy consumption; automated fault and disturbance analysis; and power quality control.Originality/valueGiven the diversity of issues reviewed, a more capability-focused research agenda is needed to further synthesize empirical findings for AI-based smart grids. Research may converge toward more focus on business rules systems, that may best support smart grid design, proof development, governance and effectiveness. These AIA technologies must be further integrated with smart grid project management methodologies and enable a greater diversity of renewable and non-renewable production sources.

Minimization of Binary Decision Diagrams for Systems of Completely Defined Boolean Functions using Algebraic Representations of Cofactors

In design systems for digital VLSI (very large integrated circuits), the BDD is used for VLSI verification, as well as for technologically independent optimization, performed as the first stage in the synthesis of logic circuits in various technological bases. BDD is an acyclic graph defining a Boolean function or a system of Boolean functions. Each vertex of this graph corresponds to the complete or reduced Shannon expansion formula. Having constructed BDD representation for systems of Boolean functions, it is possible to perform additional logical optimization based on the proposed method of searching for algebraic representations of cofactors (subfunctions) of the same BDD level in the form of a disjunction or conjunction of other cofactors of this BDD level. The method allows to reduce the number of literals by replacing the Shannon expansion formulas with simpler formulas that are disjunctions or conjunctions of cofactors, and to reduce the number of literals in specifying a system of Boolean functions. The number of literals in algebraic multilevel representations of systems of fully defined Boolean functions is the main optimization criterion in the synthesis of combinational circuits from library logic gates.

Minimization of binary decision diagrams for systems of completely defined Boolean functions using Shannon expansions and algebraic representations of cofactors

In the systems of digital VLSI design (Very Large Integrated Circuits), the BDD (Binary Decision Diagram) is used for VLSI verification, as well as for technologically independent optimization as the first stage in the synthesis of logic circuits in various technological bases. The BDD is an acyclic graph defining a Boolean function or a system of Boolean functions. Each vertex of this graph corresponds to the complete or reduced Shannon expansion formula. When BDD representation for systems of Boolean functions is constructed, it is possible to perform additional logical optimization based on the proposed method of searching for algebraic representations of cofactors (subfunctions) of the same BDD level in the form of a disjunction, conjunction either exclusive-or of cofactors of the same level or lower levels of BDD. A directed BDD graph for a system of functions is constructed on the basis of Shannon expansion of all component functions of the system by the same permutation of variables. The method allows to reduce the number of literals by replacing the Shannon expansion formulas with simpler formulas that are disjunctions or conjunctions of cofactors, and to reduce the number of literals in specifying a system of Boolean functions. The number of literals in algebraic multilevel representations of systems of fully defined Boolean functions is the main optimization criterion in the synthesis of combinational circuits from librarian logic elements.

Building Information Modeling Methods for Post-Earthquake Retrofitting Visualization of Buildings Using Augmented Reality

The post-earthquake retrofitting and repair process of a building is a key factor in improving its seismic capability. A thorough understanding of retrofitting methods and processes will aid in repairing post-earthquake buildings and improving seismic resilience. This study aims to develop a visualization framework for the post-earthquake retrofitting of buildings which builds models based on building information modeling (BIM) and realizes visualization using augmented reality (AR). First, multi-level representation methods and coding criteria are used to process the models for a damaged member. Then, an information collection template is designed for integrating multi-dimensional information, such as damage information, retrofitting methods, technical solutions, and construction measures. Subsequently, a BIM model is presented in three dimensions (3D) using AR. Finally, the visualization process is tested through experiments, which demonstrate the feasibility of using the framework to visualize the post-earthquake retrofitting of a building.

Visualizing large-scale high-dimensional data via hierarchical embedding of KNN graphs

Visualizing intrinsic structures of high-dimensional data is an essential task in data analysis. Over the past decades, a large number of methods have been proposed. Among all solutions, one promising way for enabling effective visual exploration is to construct a k-nearest neighbor (KNN) graph and visualize the graph in a low-dimensional space. Yet, state-of-the-art methods such as the LargeVis still suffer from two main problems when applied to large-scale data: (1) they may produce unappealing visualizations due to the non-convexity of the cost function; (2) visualizing the KNN graph is still time-consuming. In this work, we propose a novel visualization algorithm that leverages a multi-level representation to achieve a high-quality graph layout and employs a cluster-based approximation scheme to accelerate the KNN graph layout. Experiments on various large-scale datasets indicate that our approach achieves a speedup by a factor of five for KNN graph visualization compared to LargeVis and yields aesthetically pleasing visualization results.

AttaNet: Attention-Augmented Network for Fast and Accurate Scene Parsing

Two factors have proven to be very important to the performance of semantic segmentation models: global context and multi-level semantics. However, generating features that capture both factors always leads to high computational complexity, which is problematic in real-time scenarios. In this paper, we propose a new model, called Attention-Augmented Network (AttaNet), to capture both global context and multi-level semantics while keeping the efficiency high. AttaNet consists of two primary modules: Strip Attention Module (SAM) and Attention Fusion Module (AFM). Viewing that in challenging images with low segmentation accuracy, there are a significantly larger amount of vertical strip areas than horizontal ones, SAM utilizes a striping operation to reduce the complexity of encoding global context in the vertical direction drastically while keeping most of contextual information, compared to the non-local approaches. Moreover, AFM follows a cross-level aggregation strategy to limit the computation, and adopts an attention strategy to weight the importance of different levels of features at each pixel when fusing them, obtaining an efficient multi-level representation. We have conducted extensive experiments on two semantic segmentation benchmarks, and our network achieves different levels of speed/accuracy trade-offs on Cityscapes, e.g., 71 FPS/79.9% mIoU, 130 FPS/78.5% mIoU, and 180 FPS/70.1% mIoU, and leading performance on ADE20K as well.

Multi-level embeddings for processing Arabic social media contents

Embeddings are very popular representations that allow computing semantic and syntactic similarities between linguistic units from text co-occurrence matrix. Units can vary from character n-grams to words, including more coarse-grained units such as sentences and documents. Recently, multi-level embeddings combining representations from different units have been proposed as an alternative to single-level embeddings to account for the internal structure of words (i.e., morphology) and help systems to generalise well over out of vocabulary words. These representations, either pre-trained or learned, have shown to be quite effective, outperforming word-level baselines in several NLP tasks such as machine translation, part of speech tagging and named entity recognition. Our aim here is to contribute to this line of research proposing for the first time in Arabic NLP an in-depth study of the impact of various subwords configurations ranging from character to character n-grams (including word) for social media text classification. We propose several neural architectures to learn character, subword and word embeddings, as well as a combination of these three levels, exploring different composition functions to obtain the final representation of a given text. To evaluate the effectiveness of these representations, we perform extrinsic evaluations on three text classification tasks (sentiment analysis, emotion detection and irony detection) while accounting for different Arabic varieties (Modern Standard Arabic, dialects (Levantine and Maghrebi)). For each task, we experiment with well-known dialect-agnostic and dialect-specific datasets, including those that have been recently used in shared tasks to better compare our results with those reported in previous studies on the same datasets. The results show that the multi-level embeddings we propose outperform current static and contextualised embeddings as well as best performing state of the art models in sentiment and emotion detection. In addition, we achieve competitive results in irony detection. Our models are also the most productive across dialects observing that different dialects require different composition configurations. We finally show that these performances tend to increase when coupling the multi-level representations with task-specific features.

Logical minimization for combinatorial structure in FPGA

The paper describes the research results of application efficiency of minimization programs of functional descriptions of combinatorial logic blocks, which are included in digital devices projects that are implemented in FPGA. Programs are designed for shared and separated function minimization in a disjunctive normal form (DNF) class and minimization of multilevel representations of fully defined Boolean functions based on Shannon expansion with finding equal and inverse cofactors. The graphical form of such representations is widely known as binary decision diagrams (BDD). For technological mapping the program of "enlargement" of obtained Shannon expansion formulas was applied in a way that each of them depends on a limited number of k input variables and can be implemented on one LUT-k – a programmable unit of FPGA with k input variables. It is shown that a preliminary logic minimization, which is performed on the domestic programs, allows improving design results of foreign CAD systems such as Leonardo Spectrum (Mentor Graphics), ISE (Integrated System Environment) Design Suite and Vivado (Xilinx). The experiments were performed for FPGA families’ Virtex-II PRO, Virtex-5 and Artix-7 (Xilinx) on standard threads of industrial examples, which define both DNF systems of Boolean functions and systems represented as interconnected logical equations.

Multilevel Mapping From Remote Sensing Images: A Case Study of Urban Buildings

Remote sensing mapping plays an important role in understanding regional development and geographical environment characteristics. Traditional remote sensing mapping at different levels usually fails to consider the shape, quantity, distribution, and position features of map objects. Therefore, a multilevel representation of urban buildings is realized based on the proposed framework for multilevel mapping from remote sensing images. In this process, the Mask R-CNN method is first applied to extract buildings from remote sensing images. Then, the orthogonal shape features of the extracted buildings are reconstructed based on corner detection, and urban roads are generated by extracting the internal structural characteristics of urban buildings for further multilevel representation. Finally, three innovative raster-based generalization algorithms, including simplification, aggregation, and typification based on Hough line detection technology, are developed for a multilevel representation of urban buildings. The experimental results reveal that the proposed methods can effectively realize multilevel mapping of urban buildings from remote sensing images while meeting basic cartographic requirements.

Direct 3D printing of multi-level voxel models

Abstract We present a direct method for the additive manufacturing of multi-level voxelized models that achieves a better surface finish. We have developed a new multi-level marching squares algorithm to identify the boundary of the multi-level voxelized model. We have also developed methods to use the multi-level voxelization to perform the infill operation based on user-defined infill density. We directly generate the GCode that is input into the 3D printer for printing. Our method overcomes the issues associated with the slicing operation for standard CAD models. In addition, we can directly print thresholded voxel models that are output from CT or MRI scans to get a physical 3D representation of medical data. We show that our method performs well by directly printing test models of multi-level voxel representation of complex CAD geometries, and cardiac CT data.

Hierarchical, Dense and Dynamic 3D Reconstruction Based on VDB Data Structure for Robotic Manipulation Tasks.

This paper presents a novel approach to implement hierarchical, dense and dynamic reconstruction of 3D objects based on the VDB (Variational Dynamic B + Trees) data structure for robotic applications. The scene reconstruction is done by the integration of depth-images using the Truncated Signed Distance Field (TSDF). The proposed reconstruction method is based on dynamic trees in order to provide similar reconstruction results to the current state-of-the-art methods (i.e., complete volumes, hashing voxels and hierarchical volumes) in terms of execution time but with a direct multi-level representation that remains real-time. This representation provides two major advantages: it is a hierarchical and unbounded space representation. The proposed method is optimally implemented to be used on a GPU architecture, exploiting the parallelism skills of this hardware. A series of experiments will be presented to prove the performance of this approach in a robot arm platform.

A deep hybrid neural network for single image dehazing via wavelet transform

Image dehazing is a fast-growing research area in image processing and computer vision. Due to the extreme fog, haze, and air dispersion within an environment, the hazy image raises several challenges in retrieving original image information type. However, past techniques endure massive computation complexity and even the distortion of original images such as halos and over-saturation. In this research work, a new wavelet Hybrid (Local-Global Combined) Network is proposed for single image dehazing using a convolution neural network (CNN) in the wavelet domain (WH-Net). It is observed that low-level features such as edges are more important than high-level features such as texture. So we have used 2-DWT to decompose the single image model into the frequency subbands, which performs more quickly. It is demonstrated that the estimation of wavelet sub-bands reformulates the trainable end-to-end learning with a special architecture where DWT and IDWT are the feature extraction layers instead of Conv and Deconv, distinguishing it from classical CNN networks. The WHNet method is designed to achieve multi-level representations of hazy images to provide local and global information. The proposed network prominent features are designed with fewer convolution layers without decreasing performance relative to the commonly observed deeper learning models. Compared to several state-of-the-art algorithms, our proposed WHNet neural network surpasses in visual and quantitative performances on three public datasets.

Using association rules mining, multi-level data representation and cross-layer operations for data management in mobile ad hoc network

Using association rules mining, multi-level data representation and cross-layer operations for data management in mobile ad hoc network

Hybrid-COVID: a novel hybrid 2D/3D CNN based on cross-domain adaptation approach for COVID-19 screening from chest X-ray images.

The novel Coronavirus disease (COVID-19), which first appeared at the end of December 2019, continues to spread rapidly in most countries of the world. Respiratory infections occur primarily in the majority of patients treated with COVID-19. In light of the growing number of COVID-19 cases, the need for diagnostic tools to identify COVID-19 infection at early stages is of vital importance. For decades, chest X-ray (CXR) technologies have proven their ability to accurately detect respiratory diseases. More recently, with the availability of COVID-19 CXR scans, deep learning algorithms have played a critical role in the healthcare arena by allowing radiologists to recognize COVID-19 patients from their CXR images. However, the majority of screening methods for COVID-19 reported in recent studies are based on 2D convolutional neural networks (CNNs). Although 3D CNNs are capable of capturing contextual information compared to their 2D counterparts, their use is limited due to their increased computational cost (i.e. requires much extra memory and much more computing power). In this study, a transfer learning-based hybrid 2D/3D CNN architecture for COVID-19 screening using CXRs has been developed. The proposed architecture consists of the incorporation of a pre-trained deep model (VGG16) and a shallow 3D CNN, combined with a depth-wise separable convolution layer and a spatial pyramid pooling module (SPP). Specifically, the depth-wise separable convolution helps to preserve the useful features while reducing the computational burden of the model. The SPP module is designed to extract multi-level representations from intermediate ones. Experimental results show that the proposed framework can achieve reasonable performances when evaluated on a collected dataset (3 classes to be predicted: COVID-19, Pneumonia, and Normal). Notably, it achieved a sensitivity of 98.33%, a specificity of 98.68% and an overall accuracy of 96.91%

Using Function Allocation Methods to Inform Research Roadmaps: A Human-Robot Interaction Case Study for Medjool Date Thinning

The objective of the work is to develop a roadmap to support robotic Medjool date thinning system (RDTS) human-robot interaction research. The focus of the work relates to the allocation of functions among the human operator and the envisioned RDTS. We reviewed the current status of enabling technologies. We used the abstraction hierarchy multilevel representation framework to identify the physical functions and resources of the envisioned system. By considering the current status and the abstraction hierarchy, we developed four technically possible function allocation options. These analyses formed the basis for discussion to support research roadmap refinement. The researchers are using this refined roadmap to align their goals to ensure a better RDTS research outcome.