Feature Hierarchy Research Articles

AHM-FC Model in the Evolution Analysis of Ideological and Political Subjects

Abstract This paper proposes a new fusion neural network (FC) based on attribute hierarchy (AHM). This paper studies the evolution model of ideological and political subjects based on this method. It is abbreviated as the AHM-FC model with unequal weighted features. Then, based on the Bayesian network and least squares feature hierarchy analysis, this paper gives the conditional probability and attribute weight of ideological and political subjects. Furthermore, this paper establishes the theme evolution model of “double fusion and three stages.” In this way, it has been fully applied in ideological and political teaching in colleges and universities. Experimental results show that FC-AHM based on unequal weights can predict the evolution of ideological and political objects with higher accuracy.

Classification of semantic segmentation using fully convolutional networks based unmanned aerial vehicle application

The classification of semantic segmentation-based unmanned aerial vehicle (UAV) application based on the datasets used in this work and the necessary data preprocessing steps for the optimization and implementation of the models are also involved. The optimization of the various models was done using the evaluation metrics and loss functions because deep neural networks (DNNs) are just about writing a cost function and its subsequent optimization. convolutional neural network (CNN) is a common type of artificial neural network (ANN) that has found application in numerous tasks, such as image and video recognition, image classification, recommender systems, financial time series, medical image analysis, and natural language processing. CNN is developed to automatically and adaptively learn spatial feature hierarchies via backpropagation using numerous building blocks, such as pooling, convolution, and fully connected layers. The result of identification was excellent. The image segmentation was detected and comprehend the actual components of an image down to the pixel level. The result created an entire image segmentation masks with instances using the new label editor in the label box.

High-Level Semantic Interpretation of the Russian Static Models Structure

Since its inception, the Word2vec vector space has become a universal tool both for scientific and practical activities. Over time, it became clear that there is a lack of new methods for interpreting the location of words in vector spaces. The existing methods included consideration of analogies or clustering of a vector space. In recent years, an approach based on probing—analysis of the impact of small changes in the model on the result—has been actively developed. In this paper, we propose a new method for interpreting the arrangement of words in a vector space, applicable for the high-level interpretation of the entire space as a whole. The method provides for identifying the main directions which are selecting large groups of words (about a third of all the words in the model’s dictionary) and opposing them by some semantic features. The method allows us to build a shallow hierarchy of such features. We conducted our experiments on three models trained in different corpora: Russian National Corpus, Araneum Russicum and a collection of scientific articles from different subject domains. For our experiments, we used only nouns from the models’ dictionaries. The article considers an expert interpretation of such division up to the third level. The set of selected features and their hierarchy differ from model to model, but they have a lot in common. We have found that the identified semantic features depend on the texts comprising a corpus used for the model training, their subject domain, and style. The resulting division of words does not always correlate with the common sense used for ontology development. For example, one of the coinciding features is the abstract or material nature of the object. However, at the upper level of models, words are divided into everyday/special lexis, archaic lexis, proper names and common nouns. The article provides examples of words included in the derived groups.

CE-454610-4 A CASE OF INAPPROPRIATE CLASSIFICATION OF VENTRICULAR FIBRILLATION AS SINUS TACHYCARDIA

Accurate differentiation between SVT and malignant ventricular tachyarrhythmias in ICDs is augmented with the use of device specific SVT discriminators. However, these enhancements rely on nonprogrammable algorithms that on occasion may be responsible for inappropriate withholding of VT/VF therapy. It is important to understand the reason for rhythm misclassifications so that opportunities for improvement in ICD programming can be identified. N/A A 46-year-old man with ischemic cardiomyopathy, severe left ventricular dysfunction, left bundle branch block, prior VF arrest, AF on amiodarone therapy with a Medtronic Claria CRT-D suffered a sudden cardiac arrest at home. ICD parameter settings are noted in Table. ICD electrograms showed initial monomorphic VT with cycle length 340 ms which was treated with burst anti-tachycardia pacing that resulted in a slower VT below the VT detection rate of 162 bpm. This was sustained for 5 minutes until the rhythm devolved to VF and met the criteria for VT/VF event using the combined count number of intervals to detect (CCNID, Figure). SVT discrimination was applied as the median V-V interval was greater than the programmed SVT limit of 260 ms due to occasional undersensing. The SVT discrimination feature is designed to confirm 1:1 AV conduction using pre-specified criteria. The algorithm then relies on a simultaneous sinus tachycardia (ST) count that is established using an analysis of expected RR intervals and successive PR intervals concurrently calculated. A minimum ST count of 6 was fulfilled leading to classification of the event as sinus tachycardia and withholding for VT/VF therapy (Figure). Hierarchy of sinus tachycardia in the SVT discrimination logic withholds application of waveform analysis. Criteria for VT/VF event was met 1 minute later again based on CCNID. A 36J shock was delivered unsuccessfully terminating the rhythm. Ongoing VF was undersensed in the presence of high output post shock pacing. This case illustrates several failures of tachyarrhythmia detection leading to a tragic failure of ICD therapy. Algorithm features such as the minimum requirements for 1:1 AV conduction, sinus tachycardia counter, and hierarchy of the SVT discrimination feature to withhold waveform analysis led to inappropriate withholding of therapy for VF. Understanding elements of these complex algorithms may be helpful for individualizing device program and in future development of additional SVT discrimination enhancements.

Decision trees, entropy, and the contrastive feature hierarchy

Dresher (2009) argues that language-particular hierarchies of features are the best way to identify contrastive features in a phonological inventory. While not universal, this ordering of features is also not fully unconstrained. But what limits the space of possible feature orders remains an open question. This paper demonstrates how the concept of entropy establishes a partial ordering of features that both allows for but also constrains language-particular variation. Specifically, a decision tree machine learning algorithm is employed to dynamically impose structure on the hypothesis space of possible feature orders.

CervixFuzzyFusion for cervical cancer cell image classification

Cervical cancer is a common type of tumor that occurs in the cervix. The cervical cells in the cervix contain millions of cells with various orientations and overlaps. It is an extensive process to segment and annotate the cytoplasm and nuclei from the unsegmented cell images for better classification. In this paper, we propose an automated computerized system to classify unsegmented cervical cell images, which is achieved by using convolutional neural networks (CNN) and vision transformer (ViT) models. CNN automatically learns the spatial hierarchy of features, improving medical image classification. ViT captures long-range dependencies in extensive image recognition applications with a sophisticated encoder and global self-attention mechanisms. A novel cervix feature fusion method (CFF) that fuses the features of the pre-trained DenseNet201 and vision transformer: shifted patch tokenization (SPT) and locality self-attention (LSA) models. This fusion helps to get both local and global features from the cervical cell images. The fuzzy feature selection (FFS) method is used to select discriminative features from the fused feature vector for better classification of the cell abnormalities. The proposed method uses unsegmented cervical cell images from the publicly available SIPaKMeD dataset. The accuracy of the proposed model achieved 96.13% greater accuracy than the state-of-the-art methods despite having a smaller dataset for unsegmented cervical cell images.

NeuroSeg-II: A deep learning approach for generalized neuron segmentation in two-photon Ca2+ imaging.

The development of two-photon microscopy and Ca2+ indicators has enabled the recording of multiscale neuronal activities in vivo and thus advanced the understanding of brain functions. However, it is challenging to perform automatic, accurate, and generalized neuron segmentation when processing a large amount of imaging data. Here, we propose a novel deep-learning-based neural network, termed as NeuroSeg-II, to conduct automatic neuron segmentation for in vivo two-photon Ca2+ imaging data. This network architecture is based on Mask region-based convolutional neural network (R-CNN) but has enhancements of an attention mechanism and modified feature hierarchy modules. We added an attention mechanism module to focus the computation on neuron regions in imaging data. We also enhanced the feature hierarchy to extract feature information at diverse levels. To incorporate both spatial and temporal information in our data processing, we fused the images from average projection and correlation map extracting the temporal information of active neurons, and the integrated information was expressed as two-dimensional (2D) images. To achieve a generalized neuron segmentation, we conducted a hybrid learning strategy by training our model with imaging data from different labs, including multiscale data with different Ca2+ indicators. The results showed that our approach achieved promising segmentation performance across different imaging scales and Ca2+ indicators, even including the challenging data of large field-of-view mesoscopic images. By comparing state-of-the-art neuron segmentation methods for two-photon Ca2+ imaging data, we showed that our approach achieved the highest accuracy with a publicly available dataset. Thus, NeuroSeg-II enables good segmentation accuracy and a convenient training and testing process.

Automatic modulation recognition of radar signals based on histogram of oriented gradient via improved principal component analysis

Automatic modulation recognition (AMR) of radar signals plays a critical role in electronic reconnaissance. Current AMR algorithms are mainly based on convolutional neural networks (CNN), which can learn the feature hierarchy by establishing high-level features from low-level features. However, for time–frequency analysis-based methods, distinct low-level features in the time–frequency spectrum can already reflect modulation characteristics. Thus, this study develops a novel approach based on low-level shape descriptors via histograms of oriented gradients (HOG) and support vector machine (SVM). Comparison studies with classic CNN-based methods have also been done to reveal the superiority of the designed approach. Experimental results demonstrate that the HOG-SVM approach has a more efficient performance. To further enhance the classification precision under low signal-to-noise ratios, an improved principal component analysis denoising algorithm is developed to improve signal quality under intense noise background. Experiments based on simulated and measured signals demonstrate that the proposed algorithm can accurately distinguish signals under intense noise environments.

HEADSS: HiErArchical Data Splitting and Stitching software for non-distributed clustering algorithms

The increase in data volume is challenging the suitability of non-distributed and non-scalable algorithms, despite advancements in hardware. An example of this challenge is clustering. Considering that optimal clustering algorithms scale poorly with increased data volume or are intrinsically non-distributed, accurate clustering of large datasets is increasingly resource-heavy, relying on substantial and expensive compute nodes. This scenario forces users to choose between accuracy and scalability. In this work, we introduce HiErArchical Data Splitting and Stitching (HEADSS), a Python package designed to facilitate clustering at scale. By automating the splitting and stitching, it allows repeatable handling, and removal, of edge effects. We implement HEADSS in conjunction with HDBSCAN, where we achieve orders of magnitude reduction in single node memory requirements for both non-distributed and distributed implementations, with the latter offering similar order of magnitude reductions in total run times while recovering analogous accuracy. Furthermore, our method establishes a hierarchy of features by using a subset of clustering features to split the data.11Source code and examples are available at https://github.com/D-Crake/HEADSS.

A real-time semantic segmentation model using iteratively shared features in multiple sub-encoders

Recent studies show a significant growth in semantic segmentation. However, many semantic segmentation models still have a large number of parameters, making them unsuitable for resource-constrained embedded devices. To address this issue, we propose an efficient Shared Feature Reuse Segmentation (SFRSeg) model containing several novelties: a new yet effective shared-branch multiple sub-encoders design, a context mining module and a semantic aggregating module for better context granularity. In particular, our shared-branch approach improves the entire feature hierarchy by sharing the spatial and context knowledge in both shallow and deep branches. After every shared point in each sub-encoder, a proposed cascading context mining (CCM) module is deployed to filter out the noisy spatial details from the feature maps and provides a diverse size of receptive fields for capturing the latent context between multi-scale geometric shapes in the scene. To overcome the gradient vanishing issue at the early stage, we reduce the number of layers in the first sub-encoder and employ a unique multiple sub-encoders design which reprocesses the rich global feature maps through multiple sub-encoders for better feature refinement. Later, the rich semantic features generated by the efficient sub-encoders at different levels are fused by the proposed Hybrid Path Attention Semantic Aggregation (HPA-SA) module that effectively reduces the semantic gap between feature maps at different levels and alleviate the well-known boundary degeneration effect. To make it computationally efficient for resource-constrained embedded devices, a series of lightweight methods such as a lightweight encoder, a squeeze-and-excitation design, separable convolution filters, channel reduction (CR) are carefully exploited. With an exceptional performance on Cityscapes (70.6% test mIoU) and CamVid (74.7% test mIoU) data sets, the proposed model is shown to be superior over existing light real-time semantic segmentation models whilst having only 1.6 million parameters.

Theoretical Understanding of Convolutional Neural Network: Concepts, Architectures, Applications, Future Directions

Convolutional neural networks (CNNs) are one of the main types of neural networks used for image recognition and classification. CNNs have several uses, some of which are object recognition, image processing, computer vision, and face recognition. Input for convolutional neural networks is provided through images. Convolutional neural networks are used to automatically learn a hierarchy of features that can then be utilized for classification, as opposed to manually creating features. In achieving this, a hierarchy of feature maps is constructed by iteratively convolving the input image with learned filters. Because of the hierarchical method, higher layers can learn more intricate features that are also distortion and translation invariant. The main goals of this study are to help academics understand where there are research gaps and to talk in-depth about CNN’s building blocks, their roles, and other vital issues.

Spatiotemporal Feature Hierarchy-Based Blind Prediction of Natural Video Quality via Transfer Learning

In this paper, we propose a pyramidal spatiotemporal feature hierarchy (PSFH)-based no-reference (NR) video quality assessment (VQA) method using transfer learning. First, we generate simulated videos by a generative adversarial network (GAN)-based image restoration model. The residual maps between the distorted frames and simulated frames, which can capture rich information, are utilized as one input of the quality regression network. Second, we use 3D convolution operations to construct a PSFH network with five stages. The spatiotemporal features incorporating the shared features transferred from the pretrained image restoration model are fused stage by stage. Third, with the guidance of the transferred knowledge, each stage generates multiple feature mapping layers that encode different semantics and degradation information using 3D convolution layers and gated recurrent units (GRUs). Finally, five approximate perceptual quality scores and a precise prediction score are obtained by fully connected (FC) networks. The whole model is trained under a finely designed loss function that combines pseudo-Huber loss and Pearson linear correlation coefficient (PLCC) loss to improve the robustness and prediction accuracy. According to the extensive experiments, outstanding results can be obtained compared with other state-of-the-art methods. Both the source code and models are available online. <xref ref-type="fn" rid="fn1" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</xref>

Flood vulnerability assessment of urban buildings based on integrating high-resolution remote sensing and street view images

Urban flood risk management requires an extensive investigation of the vulnerability characteristics of buildings. Large-scale field surveys usually cost a lot of time and money, while satellite remote sensing and street view images can provide information on the tops and facades of buildings respectively. Thereupon, this paper develops a building vulnerability assessment framework using remote sensing and street view features. Specifically, a UNet-based semantic segmentation model, FSA-UNet (Fusion-Self-Attention-UNet) is proposed to integrate remote sensing and street view features and the vulnerability information contained in the images is fully exploited. And the building vulnerability index is generated to provide the spatial distribution characteristics of urban building vulnerability. The experiment shows that the mIoU of the proposed model can reach 82% for building vulnerability classification in Hefei, China, which is more accurate than the traditional semantic segmentation models. The results indicate that the integration of street view and remote sensing image features can improve the ability of building vulnerability assessment, and the model proposed in this study can better capture the correlation features of multi-angle images through the self-attention mechanism and combines hierarchy features and edge information to improve the classification effect. This study can support for disaster management and urban planning.

A Contour Line Group Simplification Method Based on Classified Terrain Features

Contour line group simplification methods can effectively preserve terrain features during map making and producing. This process involves two main steps, namely terrain feature line extraction and contour bend selection. The terrain feature line extraction includes two steps, that is, terrain feature point extraction and classification, and terrain feature line connection. However, to date, many similar studies have not considered the hierarchy of terrain features. Therefore, we proposed a group simplification method for contour lines based on classified terrain features and tested it on a study area with mainly positive landforms. In accordance with geomorphological theory, we divided valleys into either gradually descending or ordinary. Valley points were extracted based on the constrained Delaunay triangulation method and then classified into the two categories. Gradually descending and ordinary valley lines were then connected. The contour bends were grouped based on the valley lines extracted and then selected according to the geometric indicators of the bend group. The results have demonstrated that the valley lines extracted closely matched human perception in integrity and structure. Contour lines simplified by our method achieved effective reduction of map information and adequate retention of the main terrain structures, which are similar to those from manual simplification.

FabNet: A Features Agglomeration-Based Convolutional Neural Network for Multiscale Breast Cancer Histopathology Images Classification.

The definitive diagnosis of histology specimen images is largely based on the radiologist's comprehensive experience; however, due to the fine to the coarse visual appearance of such images, experts often disagree with their assessments. Sophisticated deep learning approaches can help to automate the diagnosis process of the images and reduce the analysis duration. More efficient and accurate automated systems can also increase the diagnostic impartiality by reducing the difference between the operators. We propose a FabNet model that can learn the fine-to-coarse structural and textural features of multi-scale histopathological images by using accretive network architecture that agglomerate hierarchical feature maps to acquire significant classification accuracy. We expand on a contemporary design by incorporating deep and close integration to finely combine features across layers. Our deep layer accretive model structure combines the feature hierarchy in an iterative and hierarchically manner that infers higher accuracy and fewer parameters. The FabNet can identify malignant tumors from images and patches from histopathology images. We assessed the efficiency of our suggested model standard cancer datasets, which included breast cancer as well as colon cancer histopathology images. Our proposed avant garde model significantly outperforms existing state-of-the-art models in respect of the accuracy, F1 score, precision, and sensitivity, with fewer parameters.

Semantic components of laughter behavior: a lexical field study of 14 translations ofOne Flew Over the Cuckoo’s Nest

AbstractThis paper builds on a novel methodology of lexical semantics exemplified on lexical field theory by using several translations of Ken Kesey’sOne Flew Over the Cuckoo’s Nest. The present study, a large-scale collaboration, presents and compares the results forlaugh,smile,grin, giggle, and other words for laughter behaviors across 14 languages and in extensive detail. The key results answer the question of what semantic dimensions the vocabularies of the various languages distinguish as marked by lexical contrasts and can inform future research in humor as well as translation studies. Based on our findings, a key marking emerges for audible (e.g., laugh) versus non-audible (e.g., smile) behaviors, as Indo-European vocabularies treat smiling as a less marked variant of laughing, e.g., Germanlächeln, Italiansorridere, Polishuśmiech, Turkishgülüm,but further orthogonal dimensions are documented as well, for example, aggressive, concealed, loud, or suppressed behavior. An updated hierarchy of these semantic features is proposed, and the results are presented in graphic visualizations, which also help illustrate idiosyncrasies of individual languages that go against the general trends. Exceptions to these general trends include lemmata that can cover both audible and inaudible behavior straddling what we claimed is the most important distinction (e.g., Danishgrine). Finally, we outline a probabilistic method to compare word senses across languages based on aligned corpora large enough for computational approaches.

Towards low-budget energy efficiency design in additive manufacturing based on variational scale-aware transformer

In opposite to subtractive manufacturing methodologies, additive manufacturing (AM) embraces plenty of sustainable advantages to decrease energy consumption and material waste. With the explosive data growth in AM, machine learning (ML) attracts widespread attention to enable AM designers to critically estimate energy consumption in the design stage, and thus adopt the optimal part geometry and process planning for cleaner production in AM. However, recent ML methods require massive annotated labels and usually neglect the dimensional variations of products, thereby leading to deficient energy efficiency design performance with a high experimental budget. To address these issues, the Variational Scale-aware Transformer (VST) network is proposed to precisely predict AM energy consumption with modest computational and experimental costs for low-budget energy efficiency design in AM. The task of AM energy consumption prediction is first modeled as a non-linear regression problem, where the sliced layered image (SLI) is generated as the network input in light of the layer-wise manufacturing principle in AM. The proposed VST leverages its novel multi-stage architecture to transfer the learnt comprehensive understanding from a small dimensional scale to a large dimensional scale, adapting to the increasing part dimension and geometry complexity. The multi-scale awareness module (MSAM) is further devised as a four-stage feature hierarchy for generating feature maps in different dimensional scales, efficiently enabling the seamless fusion of the fine-grained, coarse-grained, and local variational scale embeddings. Extensive numerical and physical experimental results demonstrate that the proposed method consistently outperforms the state-of-the-art networks and enhances the energy prediction performance for low-budget energy efficiency design in AM.

Residual dynamic mode decomposition: robust and verified Koopmanism

Dynamic mode decomposition (DMD) describes complex dynamic processes through a hierarchy of simpler coherent features. DMD is regularly used to understand the fundamental characteristics of turbulence and is closely related to Koopman operators. However, verifying the decomposition, equivalently the computed spectral features of Koopman operators, remains a significant challenge due to the infinite-dimensional nature of Koopman operators. Challenges include spurious (unphysical) modes and dealing with continuous spectra, which both occur regularly in turbulent flows. Residual dynamic mode decomposition (ResDMD), introduced by Colbrook &amp; Townsend (Rigorous data-driven computation of spectral properties of Koopman operators for dynamical systems. 2021. arXiv:2111.14889), overcomes such challenges through the data-driven computation of residuals associated with the full infinite-dimensional Koopman operator. ResDMD computes spectra and pseudospectra of general Koopman operators with error control and computes smoothed approximations of spectral measures (including continuous spectra) with explicit high-order convergence theorems. ResDMD thus provides robust and verified Koopmanism. We implement ResDMD and demonstrate its application in various fluid dynamic situations at varying Reynolds numbers from both numerical and experimental data. Examples include vortex shedding behind a cylinder, hot-wire data acquired in a turbulent boundary layer, particle image velocimetry data focusing on a wall-jet flow and laser-induced plasma acoustic pressure signals. We present some advantages of ResDMD: the ability to resolve nonlinear and transient modes verifiably; the verification of learnt dictionaries; the verification of Koopman mode decompositions; and spectral calculations with reduced broadening effects. We also discuss how a new ordering of modes via residuals enables greater accuracy than the traditional modulus ordering (e.g. when forecasting) with a smaller dictionary. This result paves the way for more significant dynamic compression of large datasets without sacrificing accuracy.

Deeply Explain CNN Via Hierarchical Decomposition

In computer vision, some attribution methods for explaining CNNs attempt to study how the intermediate features affect network prediction. However, they usually ignore the feature hierarchies among the intermediate features. This paper introduces a hierarchical decomposition framework to explain CNN’s decision-making process in a top-down manner. Specifically, we propose a gradient-based activation propagation (gAP) module that can decompose any intermediate CNN decision to its lower layers and find the supporting features. Then we utilize the gAP module to iteratively decompose the network decision to the supporting evidence from different CNN layers. The proposed framework can generate a deep hierarchy of strongly associated supporting evidence for the network decision, which provides insight into the decision-making process. Moreover, gAP is effort-free for understanding CNN-based models without network architecture modification and extra training processes. Experiments show the effectiveness of the proposed method. The data and source code will be publicly available at https://mmcheng.net/hdecomp/ .

Convolutional Transformer Fusion Blocks for Multi-Modal Gesture Recognition

Gesture recognition defines an important information channel in human-computer interaction. Intuitively, combining inputs from multiple modalities improves the recognition rate. In this work, we explore multi-modal video-based gesture recognition tasks by fusing spatio-temporal representation of relevant distinguishing features from different modalities. We present a self-attention based transformer fusion architecture to distill the knowledge from different modalities in two-stream convolutional neural networks (CNNs). For this, we introduce convolutions into the self-attention function and design the Convolutional Transformer Fusion Blocks (CTFB) for multi-modal data fusion. These fusion blocks can be easily added at different abstraction levels of the feature hierarchy in existing two-stream CNNs. In addition, the information exchange between two-stream CNNs along the feature hierarchy has so far been barely explored. We propose and evaluate different architectures for multi-level fusion pathways using CTFB to gain insights into the information flow between both streams. Our method achieves state-of-the-art or competitive performance on three benchmark gesture recognition datasets: a) IsoGD, b) NVGesture, and c) IPN hand. Extensive evaluation demonstrates the effectiveness of the proposed CTFB both in terms of recognition rate as well as resource efficiency.