Search

The recurrent network architecture is a widely used model in sequence modeling, but its serial dependency hinders the computation parallelization, which makes the operation inefficient. The same problem was encountered in serial adder at the early stage of digital electronics. In this paper, we discuss the similarities between recurrent neural network (RNN) and serial adder. Inspired by carry-lookahead adder, we introduce carry-lookahead module to RNN, which makes it possible for RNN to run in parallel. Then, we design the method of parallel RNN computation, and finally Carry-lookahead RNN (CL-RNN) is proposed. CL-RNN takes advantages in parallelism and flexible receptive field. Through a comprehensive set of tests, we verify that CL-RNN can perform better than existing typical RNNs in sequence modeling tasks which are specially designed for RNNs. Code and models are available at: https://github.com/WinnieJiangHW/Carry-lookahead_RNN. [ABSTRACT FROM AUTHOR]

Background and objective: Leukocyte classification and cytometry have wide applications in medical domain, previous researches usually exploit machine learning techniques to classify leukocytes automatically. However, constrained by the past development of machine learning techniques, for example, extracting distinctive features from raw microscopic images are difficult, the widely used SVM classifier only has relative few parameters to tune, these methods cannot efficiently handle fine-grained classification cases when the white blood cells have up to 40 categories. Methods: Based on deep learning theory, a systematic study is conducted on finer leukocyte classification in this paper. A deep residual neural network based leukocyte classifier is constructed at first, which can imitate the domain expert’s cell recognition process, and extract salient features robustly and automatically. Then the deep neural network classifier’s topology is adjusted according to the prior knowledge of white blood cell test. After that the microscopic image dataset with almost one hundred thousand labeled leukocytes belonging to 40 categories is built, and combined training strategies are adopted to make the designed classifier has good generalization ability. Results: The proposed deep residual neural network based classifier was tested on microscopic image dataset with 40 leukocyte categories. It achieves top-1 accuracy of 77.80%, top-5 accuracy of 98.75% during the training procedure. The average accuracy on the test set is nearly 76.84%. Conclusions: This paper presents a fine-grained leukocyte classification method for microscopic images, based on deep residual learning theory and medical domain knowledge. Experimental results validate the feasibility and effectiveness of our approach. Extended experiments support that the fine-grained leukocyte classifier could be used in real medical applications, assist doctors in diagnosing diseases, reduce human power significantly. [ABSTRACT FROM AUTHOR]

Many types of heterogeneous CAD models created by previous product development activities are stored in enterprise databases. Searching reusable CAD models for engineer design requirements is still very difficult. In this paper, we propose an ontology-based semantic retrieval approach for heterogeneous 3D CAD models by using semantic web theory and information retrieval theory. By exploiting layered feature ontology and ontology mapping, a uniform description for heterogeneous CAD models is generated as semantic descriptor. Meanwhile, to improve the representative ability, several heuristic rules are constructed with the aid of SWRL language and ontology reasoning is acted on the semantic descriptors, which makes our retrieval system achieve better performance. Furthermore, multi-mode similarity measurement is carried out to meet engineers’ diversified requirements. We demonstrate the feasibility and effectiveness of our approach through experiments on heterogeneous 3D CAD model library. [ABSTRACT FROM AUTHOR]

Design rationale (DR) is an important part of design knowledge. Effective reuse of DR depends on its successful retrieval. In this paper, an approach for structured DR retrieval using ontology-aided indexing is presented. Firstly, an ontology-based semantic model of DR is developed based on the extended IBIS-based DR representation in order to effectively utilise the semantics embedded in DR. Then, an ontology-aided indexing method is proposed to build indexes for DR records to index the semantic concepts and relationships in DR. Furthermore, three kinds of query modes are developed to support flexible querying, among which natural language input query and DR record-based query have much more semantics and thus lead to better retrieval results. Finally, a prototype system is implemented. Experimental results show the effectiveness of the proposed approach. [ABSTRACT FROM PUBLISHER]

How to quickly, accurately retrieve and effectively reuse 3D CAD models that conform to user's design intention has become an urgent problem in product design. However, there are several problems with the existing retrieval methods, like not being fast, or accurate, or hard to use. Hence it is difficult to meet the actual needs of the industry. In this paper, we propose a 3D CAD model retrieval approach that considers the speed, accuracy and ease of use at the same time, based on sketches and unsupervised learning. Firstly, the loop is used as the fundamental element of sketch/view, and automatic structural semantics capture algorithms are proposed to extract and construct attributed loop relation tree; Secondly, the recursive neural network based deep variational autoencoders is constructed and optimized to transform arbitrary shapes and sizes of loop relation tree into fixed length descriptor; Finally, based on the fixed length vector descriptor, the sketches and views of 3D CAD models are embedded into the same target feature space, and k -nearest neighbors algorithm is adopted to conduct fast CAD model matching on the feature space. In this manner, a prototype 3D CAD model retrieval system is developed. Experiments on the dataset containing about two thousand 3D CAD models validate the feasibility and effectiveness of the proposed approach. [ABSTRACT FROM AUTHOR]

Schizophrenia is a debilitating psychiatric disorder that can significantly affect a patient's quality of life and lead to permanent brain damage. Although medical research has identified certain genetic risk factors, the specific pathogenesis of the disorder remains unclear. Despite the prevalence of research employing magnetic resonance imaging, few studies have focused on the gene level and gene expression profile involving a large number of screened genes. However, the high dimensionality of genetic data presents a great challenge to accurately modeling the data. To tackle the current challenges, this study presents a novel feature selection strategy that utilizes heuristic feature fusion and a multi-objective optimization genetic algorithm. The goal is to improve classification performance and identify the key gene subset for schizophrenia diagnostics. Traditional gene screening techniques are inadequate for accurately determining the precise number of key genes associated with schizophrenia. Our innovative approach integrates a filter-based feature selection method to reduce data dimensionality and a multi-objective optimization genetic algorithm for improved classification tasks. By combining the filtering and wrapper methods, our strategy leverages their respective strengths in a deliberate manner, leading to superior classification accuracy and a more efficient selection of relevant genes. This approach has demonstrated significant improvements in classification results across 11 out of 14 relevant datasets. The performance on the remaining three datasets is comparable to the existing methods. Furthermore, visual and enrichment analyses have confirmed the practicality of our proposed method as a promising tool for the early detection of schizophrenia. [ABSTRACT FROM AUTHOR]

The advent of deep learning has provided solutions to many challenges posed by the Internet. However, efficient localization and recognition of vulgar segments within videos remain formidable tasks. This difficulty arises from the blurring of spatial features in vulgar actions, which can render them indistinguishable from general actions. Furthermore, issues of boundary ambiguity and over-segmentation complicate the segmentation of vulgar actions. To address these issues, we present the Boundary-Match U-shaped Temporal Convolutional Network (BMUTCN), a novel approach for the segmentation of vulgar actions. The BMUTCN employs a U-shaped architecture within an encoder–decoder temporal convolutional network to bolster feature recognition by leveraging the context of the video. Additionally, we introduce a boundary-match map that fuses action boundary inform ation with greater precision for frames that exhibit ambiguous boundaries. Moreover, we propose an adaptive internal block suppression technique, which substantially mitigates over-segmentation errors while preserving accuracy. Our methodology, tested across several public datasets as well as a bespoke vulgar dataset, has demonstrated state-of-the-art performance on the latter. [ABSTRACT FROM AUTHOR]

In electroencephalography (EEG)-based affective brain–computer interfaces (aBCIs), there is a consensus that EEG features extracted from different frequency bands and channels have different abilities in emotion expression. Besides, EEG is so weak and non-stationary that easily causes distribution discrepancies for EEG data collected at different times; therefore, it is necessary to explore the affective activation patterns in cross-session emotion recognition. To address these two problems, we propose a self-weighted semi-supervised classification (SWSC) model in this article for joint EEG-based cross-session emotion recognition and affective activation patterns mining, whose merits include: 1) using both the labeled and unlabeled samples from different sessions for better capturing data characteristics; 2) introducing a self-weighted variable to learn the importance of EEG features adaptively and quantitatively; and 3) mining the activation patterns including the critical EEG frequency bands and channels automatically based on the learned self-weighted variable. Extensive experiments are conducted on the benchmark SEED_IV emotional dataset and SWSC obtained excellent average accuracies of 77.40%, 79.55%, and 81.52% in three cross-session emotion recognition tasks. Moreover, SWSC identifies that the Gamma frequency band contributes the most and the EEG channels in prefrontal, left/right temporal, and (central) parietal lobes are more important for cross-session emotion recognition. [ABSTRACT FROM AUTHOR]

Traditional clinical diagnostic aid systems for medical images are facing challenges of reliability and interpretability. Artificial intelligence has the potential to bring driving changes to disease diagnosis methods through rapid traversal of medical images and efficient classification. However, the application of artificial intelligence in the field of medical image still faces challenges. Our method combines the multiple modalities of attention which consider the most discriminative part in the images. The proposed classification method is tested on the microscopic image dataset with 40 leukocyte categories, which achieves top-1 accuracy of 84.21% and top-5 accuracy of 99.44% during the testing procedure. And experiments on the dermoscopic image dataset show that our method has good generalization ability across multiple imaging modalities. [ABSTRACT FROM AUTHOR]

We present a novel active learning approach for shape cosegmentation based on graph convolutional networks (GCNs). The premise of our approach is to represent the collections of three-dimensional shapes as graph-structured data, where each node in the graph corresponds to a primitive patch of an oversegmented shape, and is associated with a representation initialized by extracting features. Then, the GCN operates directly on the graph to update the representation of each node based on a layer-wise propagation rule, which aggregates information from its neighbors, and predicts the labels for unlabeled nodes. Additionally, we further suggest an active learning strategy that queries the most informative samples to extend the initial training samples of GCN to generate more accurate predictions of our method. Our experimental results on the Shape COSEG dataset demonstrate the effectiveness of our approach. [ABSTRACT FROM AUTHOR]

The necessary step in the diagnosis of leukemia by the attending physician is to classify the white blood cells in the bone marrow, which requires the attending physician to have a wealth of clinical experience. Now the deep learning is very suitable for the study of image recognition classification, and the effect is not good enough to directly use some famous convolution neural network (CNN) models, such as AlexNet model, GoogleNet model, and VGGFace model. In this paper, we construct a new CNN model called WBCNet model that can fully extract features of the microscopic white blood cell image by combining batch normalization algorithm, residual convolution architecture, and improved activation function. WBCNet model has 33 layers of network architecture, whose speed has greatly been improved compared with the traditional CNN model in training period, and it can quickly identify the category of white blood cell images. The accuracy rate is 77.65% for Top-1 and 98.65% for Top-5 on the training set, while 83% for Top-1 on the test set. This study can help doctors diagnose leukemia, and reduce misdiagnosis rate. [ABSTRACT FROM AUTHOR]

Temporally detecting and classifying action segments in untrimmed videos are significant for many applications, especially for the detection of vulgar actions such as sucking and caressing in video platform supervision and surveillance applications. At present, vulgar action segmentation has problems such as fuzzy spatial features and complex temporal features of the video, which affect the detection accuracy. Therefore, this paper proposed an effective Adaptive receptive field U-shaped Temporal Convolutional Network (AU-TCN) for the automatic and accurate detection of vulgar actions in the video. Firstly, considering that the current temporal convolutional network has a significant effect on temporal feature extraction, AU-TCN uses the adaptive receptive field convolution kernel to solve the problem of large differences in the average duration between different types of actions in the Internet videos and then realize the temporal attention mechanism. Secondly, the U-shaped structure based on the temporal convolutional network is introduced to effectively analyze both high-level and low-level temporal features of the model, to solve the problem that the spatial features of vulgar actions are not obvious. Finally, extensive experiments on multiple data sets, including public datasets and a self-built vulgar dataset, verify the effectiveness of the proposed model. Our method achieves state-of-the-art results on the vulgar action dataset. This is of great significance to the purification of the Internet environment. [ABSTRACT FROM AUTHOR]

Performing 3-dimensional computer-aided design (3D-CAD) model classification, retrieval, and reuse is of vital importance in industrial manufacturing, as it considerably shortens the engineering development cycle and reduces development costs. Although existing 3D model classification and retrieval methods achieve satisfactory performance when operating on meshes or point clouds, they cannot be applied directly to 3D-CAD models, which are generally represented by the boundary representation (B-rep). To address this issue, and to fully exploit the topology of B-rep, a graph structure descriptor called B-rep graph is proposed to pre-process B-rep data, and to extract the precise topological and geometric features from 3D-CAD models. Meanwhile, a novel efficient neural network called FuS-GCN, based on graph convolutional networks (GCNs), is designed to handle this graph data. To better extract the graph features and to improve the effect of pooling, the self-attention mechanism and feature fusion are incorporated into the pooling layer, yielding the proposed fusion self-attention graph pooling (FuSPool) algorithm. Finally, we demonstrate the effectiveness of FuS-GCN on 3D-CAD model data, while outperforming alternative 3D shape descriptors such as point clouds, voxels, and meshes. [ABSTRACT FROM AUTHOR]

Design pattern is widely used in the software engineering field, which enables designers to reuse existing mature designs from a high level perspective. Inspired by this idea, a novel approach is proposed to extract design patterns in the CAD field. First, the characteristics for a good design pattern are analyzed and the model for representing design patterns is elaborated. Then, given a set of 3D feature-based CAD models, the corresponding extraction approach is proposed, which includes three important phases: (1) extracting reusable regions with high cohesion, low coupling and moderate complexity so as to form a relative integrated function; (2) constructing candidate design patterns by clustering reusable regions using a graph-oriented agglomerative hierarchical clustering algorithm; (3) determining the final design patterns by choosing those candidate design patterns with high frequency and sufficient information. Finally, a design pattern extraction prototype system is developed, and the experimental results are presented to demonstrate the effectiveness of the approach. [ABSTRACT FROM AUTHOR]

Neural Radiance Fields (NeRF) have achieved great success in the task of synthesizing novel views that preserve the same resolution as the training views. However, it is challenging for NeRF to synthesize high-quality high-resolution novel views with low-resolution training data. To solve this problem, we propose a zero-shot super-resolution training framework (ZS-SRT) for NeRF. This framework aims to guide the NeRF model to synthesize high-resolution novel views via single-scene internal learning rather than requiring any external high-resolution training data. Our method consists of two stages. First, we learn a scene-specific degradation mapping by performing internal learning on a pretrained low-resolution coarse NeRF. Second, we optimize a super-resolution fine NeRF by conducting inverse rendering with our mapping function so as to backpropagate the gradients from low-resolution 2D space into the super-resolution 3D sampling space. Then, we further introduce a temporal ensemble strategy in the inference phase to compensate for the scene estimation errors. Our method is featured on two points: (1) it does not consume high-resolution views or additional scene data to train super-resolution NeRF; (2) it can speed up the training process by adopting a coarse-to-fine strategy. By conducting extensive experiments on public datasets, we have qualitatively and quantitatively demonstrated the effectiveness of our method. [ABSTRACT FROM AUTHOR]

Concept factorization (CF) is an effective matrix factorization model which has been widely used in many applications. In CF, the linear combination of data points serves as the dictionary based on which CF can be performed in both the original feature space as well as the reproducible kernel Hilbert space (RKHS). The conventional CF treats each dimension of the feature vector equally during the data reconstruction process, which might violate the common sense that different features have different discriminative abilities and therefore contribute differently in pattern recognition. In this paper, we introduce an auto-weighting variable into the conventional CF objective function to adaptively learn the corresponding contributions of different features and propose a new model termed Auto-Weighted Concept Factorization (AWCF). In AWCF, on one hand, the feature importance can be quantitatively measured by the auto-weighting variable in which the features with better discriminative abilities are assigned larger weights; on the other hand, we can obtain more efficient data representation to depict its semantic information. The detailed optimization procedure to AWCF objective function is derived whose complexity and convergence are also analyzed. Experiments are conducted on both synthetic and representative benchmark data sets and the clustering results demonstrate the effectiveness of AWCF in comparison with some related models. [ABSTRACT FROM AUTHOR]

Occlusion is unavoidable in real-world applications of person re-identification (ReID). To alleviate the occlusion problem, this work proposes the detection of the occluded and visible regions of the human body by suppressing the occluded region during feature generation and matching, and enhancing the significance of the visible region. This paper introduces a novel method based on pose-guided spatial attention (PGSA) and activation-based attention (AA) called dual-attention re-identification (DAReID). DAReID consists of a mask branch and a global branch and uses ResNet-50 as the backbone network. The mask branch uses PGSA to obtain the visible and occluded regions of a person and constructs pose guided coarse labels for the occluded region through keypoints of the human body, driving the network to obtain robust local features. The global branch obtains the visual activation levels of different regions through AA, and combines this with human pose information to define weighted local distances(WLD). The WLD learning strategy is applied to drive the network to learn new and more discriminative local features. Experimental results show that DAReID achieves comparable performance on the Market1501, DukeMTMC-reID, and CUHK-03 datasets. And on the Occluded-DukeMTMC dataset, DAReID outperforms the existing methods. [ABSTRACT FROM AUTHOR]

Subspace learning aims to obtain the corresponding low-dimensional representation of high dimensional data in order to facilitate the subsequent data storage and processing. Graph-based subspace learning is a kind of effective subspace learning methods by modeling the data manifold with a graph, which can be included in the general spectral regression (SR) framework. By using the least square regression form as objective function, spectral regression mathematically avoids performing eign-decomposition on dense matrices and has excellent flexibility. Recently, spectral regression has obtained promising performance in diverse applications; however, it did not take the underlying classes/tasks correlation patterns of data into consideration. In this paper, we propose to improve the performance of spectral regression by exploring the correlation among classes with low-rank modeling. The newly formulated low-rank spectral regression (LRSR) model is achieved by decomposing the projection matrix in SR by two factor matrices which were respectively regularized. The LRSR objective function can be handled by the alternating direction optimization framework. Besides some analysis on the differences between LRSR and existing related models, we conduct extensive experiments by comparing LRSR with its full rank counterpart on benchmark data sets and the results demonstrate its superiority. [ABSTRACT FROM AUTHOR]

3D shapes and 2D images usually contain complementary information for each other, and thus joint analysis of both of them will benefit some problems existing in different domains. Leveraging the connection between 2D images and 3D shapes, it's potential to mine lacking information of one modal from the other. Stemming from this insight, we design and implement a CNN architecture to jointly analyze shapes and images even with few training data guidance. The core of our architecture is a domain adaptation algorithm, which builds up the connection between underlying feature spaces of images and shapes, then aligns and correlates the intrinsic structures therein. The proposed method facilitates the recognition and retrieval tasks. Experiments on the shape recognition tasks show that our approach has superior performance under the difficult setting: zero-shot learning and few-shot learning. We also evaluate our method on the retrieval tasks, and demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Transcriptional co-activator with PDZ-binding motif (TAZ) is a WW domain-containing protein that regulates mesenchymal differentiation and organ development. It is also a downstream effector of the Hippo signaling pathway, which has been implicated in epithelial-mesenchymal transition (EMT) and tumorigenesis. However, the molecular mechanisms underlying TAZ function in these processes in the context of osteosarcoma (OS) are not well understood. We addressed this in the present study using U2OS and HOS cell lines. We found that TAZ signaling is maintained via a previously undescribed micro (mi)RNA-dependent positive feedback loop. The miRNA miR-135b, which is directly induced by TAZ, suppressed the TAZ inhibitors large tumor suppressor 2, adenomatous polyposis coli, and glycogen synthase kinase 3β, thereby amplifying TAZ signaling and inducing EMT. Overexpression of miR-135b caused constitutive activation of TAZ, which rescued the inhibition of cell proliferation and EMT induced by TAZ knockdown. These results provide evidence that TAZ and miR-135b engage in a positive feedback loop to regulate EMT and metastasis in OS, and suggest that both factors can be therapeutic targets for OS treatment. [ABSTRACT FROM AUTHOR]

Accurate segmentation of the tumour area is crucial for the treatment and prognosis of patients with bladder cancer. However, the complex information from the MRI image poses an important challenge for us to accurately segment the lesion, for example, the high distinction among people, size of bladder variation and noise interference. Based on the above issues, we propose an MD-Unet network structure, which uses multi-scale images as the input of the network, and combines max-pooling with dilated convolution to increase the receptive field of the convolutional network. The results show that the proposed network can obtain higher precision than the existing models for the bladder cancer dataset. The MD-Unet can achieve state-of-art performance compared with other methods. [ABSTRACT FROM AUTHOR]

Most previous methods of bas-relief generation run slow, or require tuning several important parameters. These issues seriously reduce the efficiency of bas-relief modeling. We introduce a fast generation method for high-quality bas-reliefs from 3D objects based on a deep learning technique. Unlike neural networks for image tasks, the proposed network for reliefs (ReliefNet) is elaborately designed to deal with a modeling problem in the field of graphics. We design our ReliefNet and equip it with a special loss function with the aim that the network can solve the essential problem of bas-relief modeling. Our network eliminates the height gaps and maintains the rich details simultaneously. The advantage over previous methods is that our method does not require parameter tuning and is a very efficient. Once the ReliefNet has been trained, a bas-relief can be produced by one feed-forward pass of the network instantly. To demonstrate the performance and effectiveness of our method, extensive experiments on a range of 3D scenes with high resolutions and comparisons to state-of-the-art methods are conducted. • The first end-to-end trained deep neural network for relief modeling from 3D scenes. • The proposed loss function is critical for the essential problem of relief modeling. • Our network makes it very efficient to create high-quality relief from a 3D scene. • A relief dataset is first constructed to train the neural network. [ABSTRACT FROM AUTHOR]

Subspace learning represents a group of algorithms to project high dimensional data onto a low dimensional subspace in order to retain the desirable properties and simultaneously reduce the dimensionality. In graph-based subspace learning algorithms, the quality of graph affects the performance of projection matrix learning a lot. Especially when data is noisy or even grossly corrupted, we cannot guarantee that the constructed graph can accurately depict the inner structure of data. Additionally, the widely used two-stage paradigm of learning the projection matrix on a given graph isolates the connection between these two stages. In this paper, we propose a general framework to get rid of these disadvantages by the inspiration of low-rank matrix recovery and spectral regression-based subspace learning. Concretely, by jointly optimizing the objectives of low-rank representation and spectral regression, on one hand we can automatically get the recovered data from noise and the graph affinity matrix and on the other hand we can perform the projection matrix learning. The formulated joint low-rank and subspace learning (JLRSL) framework can be efficiently optimized by the augmented Lagrange multiplier method. We evaluate the performance of JLRSL by conducting extensive experiments on representative benchmark data sets and the results show that the low-rank learning can greatly facilitate the process of subspace learning, leading to robust feature extraction. Moreover, comparison with the state-of-the-arts is performed. [ABSTRACT FROM AUTHOR]