Search

Graph convolutional neural networks (GCNs) are emerging neural networks for graph structures that include large features associated with each vertex. The operations of GCN can be divided into two phases - aggregation and combination. While the combination just performs matrix multiplications using trained weights and aggregated features, the aggregation phase requires graph traversal to collect features from adjacent vertices. Even though neural network applications rely on GPU’s massively parallel processing, GCN aggregation kernels exhibit rather low performance since graph processing using compressed graph structures provokes frequent irregular accesses in GPUs. In order to investigate the performance hurdles of GCN aggregation on GPU, we perform an in-depth analysis of the aggregation kernels using real GPU hardware and a cycle-accurate GPU simulator. We first analyze the characteristics of the popular graph datasets used for GCN studies. We reveal the fractions of non-zero elements in feature vectors are diverse among datasets. Based on the observation, we build two types of aggregation kernels that handle uncompressed and compressed feature vectors. Our evaluation exhibits the performance of aggregation can be significantly influenced by kernel design approaches and feature density. We also analyze the individual loads that access the data arrays of the aggregation kernels to specify critical loads. Our analysis reveals the performance of GPU memory hierarchy is influenced by access patterns and feature size of graph datasets. Based on our observations we discuss possible kernel design approaches and architectural ideas that can improve the performance of GCN aggregation.

In order to automatically generate multiple labels for the event text of the Chinese government hotline, this paper propose a multi-label classification framework based on graph convolutional network (GCN), BERT, and memory network. The framework consists of three modules: label count prediction module, label semantic insert module, and label selection module. In the label count prediction module, this paper constructs the event graph with the abstract meaning representation (AMR) and extract the event topic information vector with GCN. To predict the label count, this paper first use BERT to extract the event semantic information vector and then fuse it with the event topic information vector (GCN-BERT fusion vector) with a dynamic fusion gate. In the label semantic insert module, to obtain the event label candidate set, this paper uses a multi-hop memory network to store the event label semantic information, and then use the answer selection framework, which matches the GCN-BERT fusion vector with the event label semantic memory vector. In label selection module, this paper uses the label count based multi-label selection to sort the event label candidate set and guide to output the optimal multi-label set of the event. Comparison experimental results show that the proposed framework outperforms all baselines and ablation studies demonstrate the effectiveness of each module.

Context-based GCNs have achieved relatively good effectiveness in the sentiment analysis task, especially aspect-level sentiment analysis (ALSA). However, the previous context-based GCNs for ALSA often used GCNs with the following limitations: (i) Using GCNs limited to a few layers (two or three) due to the vanishing gradient, limiting their performance. (ii) Not considering helpful information about the hidden context between the words. To solve these limitations, this paper proposes a novel CNN over the BERT-GCN model for ALSA. The contributions of the proposed method are summarized as follows: (i) Handling the disadvantage of limiting the GCN to a few layers by adding convolutional layers of the convolutional neural network (CNN) model after GCN layers. (ii) Considering further helpful information about the hidden context between the words by combining the Bidirectional Encoder Representations from Transformers (BERT) and Bidirectional Long Short Term Memory (BiLSTM) models. The proposed model includes the following steps: First, words in sentences are converted vectors using BERT. Second, the contextualized word representations are created based on BiLSTM over word vectors. Third, significant features are extracted and represented using the GCN model with multiple convolutional layers over the contextualized word representations. Finally, the aspect-level sentiments are classified using the CNN model over the feature vectors. Experiments on three benchmark datasets illustrate that our proposed model has improved the performance of the previous context-based GCN methods for ALSA.

The problem of image segmentation is one of the most significant ones in computer vision. Recently, deep-learning methods have dominated state-of-the-art solutions that automatically or interactively divide an image into subregions. However, the limitation of deep-learning approaches is that they require a substantial amount of training data, which is costly to prepare. An alternative solution is semi-supervised image segmentation. It requires rough denotations to define constraints that are next generalized to precisely delimit relevant image regions without using train examples. Among semi-supervised strategies for image segmentation, the leading are graph-based techniques that define image segmentation as a result of pixel or region affinity graph partitioning. This paper revisits the problem of graph-based image segmentation. It approaches the problem as semi-supervised node classification in the SLIC superpixels region adjacency graph using a graph convolutional network (GCN). The performance of both spectral and spatial graph convolution operators is considered, represented by Chebyshev convolution operator and GraphSAGE respectively. The results of the proposed method applied to binary and multi-label segmentation are presented, numerically assessed, and analyzed. In its best variant, the proposed method scored the average DICE of 0.86 in the binary segmentation task and 0.79 in the multi-label segmentation task. Comparison with state-of-the-art graph-based methods, including Random Walker and GrabCut, shows that graph convolutional networks can represent an attractive alternative to the existing solutions to graph-based semi-supervised image segmentation.

In the recent era, graph neural networks are widely used on vision-to-language tasks and achieved promising results. In particular, graph convolution network (GCN) is capable of capturing spatial and semantic relationships needed for visual question answering (VQA). But, applying GCN on VQA datasets with different subtasks can lead to varying results. Also, the training and testing size, evaluation metrics and hyperparameter used are other factors that affect VQA results. These, factors can be subjected into similar evaluation schemes in order to obtain fair evaluations of GCN based result for VQA. This study proposed a GCN framework for VQA based on fine tune word representation to solve handle reasoning type questions. The framework performance is evaluated using various performance measures. The results obtained from GQA and VQA 2.0 datasets slightly outperform most existing methods. [ABSTRACT FROM AUTHOR]

Temperature is an important part of meteorological factors, which are affected by local and surrounding meteorological factors. Aiming at the problems of significant prediction error and insufficient extraction of spatial features in current temperature prediction research, this research proposes a temperature prediction model based on the Graph Convolutional Network (GCN) and Bidirectional Long Short-Term Memory (BiLSTM) and studies the influence of temperature time-series characteristics, urban spatial location, and other meteorological factors on temperature change in the study area. In this research, multi-meteorological influencing factors and temperature time-series characteristics are used instead of single time-series temperature as influencing factors to improve the time dimension of the input data through time-sliding windows. Meanwhile, considering the influence of meteorological factors in the surrounding area on the temperature change in the study area, we use GCN to extract the urban geospatial location features. The experimental results demonstrate that our model outperforms other models and has the smallest root mean squared error (RMSE) and mean absolute error (MAE) in the following 14-day and multi-region temperature forecasts. It has higher accuracy in areas with stable temperature fluctuations and small temperature differences than in baseline models. [ABSTRACT FROM AUTHOR]

The near-surface temperature prediction (NTP) is an important spatial–temporal forecast problem, which can be used to prevent temperature crises. Most of the previous approaches fail to explicitly model the long- and short-range spatial correlations simultaneously, which is critical to making an accurate temperature prediction. In this study, both long- and short-range spatial correlations are captured to fill this gap by a novel convolution operator named Long- and Short-range Convolution (LS-Conv). The proposed LS-Conv operator includes three key components, namely, Node-based Spatial Attention (NSA), Long-range Adaptive Graph Constructor (LAGC), and Long- and Short-range Integrator (LSI). To capture long-range spatial correlations, NSA and LAGC are proposed to evaluate node importance aiming at auto-constructing long-range spatial correlations, which is named as Long-range aware Graph Convolution Network (LR-GCN). After that, the Short-range aware Convolution Neural Network (SR-CNN) accounts for the short-range spatial correlations. Finally, LSI is proposed to capture both long- and short-range spatial correlations by intra-unifying LR-GCN and SR-CNN. Upon the proposed LS-Conv operator, a new model called Long- and Short-range for NPT (LS-NTP) is developed. Extensive experiments are conducted on two real-world datasets and the results demonstrate that the proposed method outperforms state-of-the-art techniques. The source code is available on GitHub: https://github.com/xuguangning1218/LS_NTP. • We jointly incorporate long- and short-range spatial correlations in NTP prediction. • Theoretical findings proved that the proposed LS-Conv is a general version of CNN. • We develop a new spatial–temporal model named LS-NTP for the NTP task. • Extensive experiments are conducted on two real-world datasets. [ABSTRACT FROM AUTHOR]

ALSC (Aspect-level Sentiment Classification) is a fine-grained task in the field of NLP (Natural Language Processing) which aims to identify the sentiment toward a given aspect. In addition to exploiting the sentence semantics and syntax, current ALSC methods focus on introducing external knowledge as a supplementary to the sentence information. However, the integration of the three categories of information is still challenging. In this paper, a novel method is devised to effectively combine sufficient semantic and syntactic information as well as use of external knowledge. The proposed model contains a sentence encoder, a semantic learning module, a syntax learning module, a knowledge enhancement module, an information fusion module and a sentiment classifier. The semantic information and syntactic information are respectively extracted via a self-attention network and a graphical convolutional network. Specifically, the KGE (Knowledge Graph Embedding) is employed to enhance the feature representation of the aspect. Then, the attention-based gate mechanism is taken to fuse three types of information. We evaluated the proposed model on three benchmark datasets and the experimental results establish strong evidence of high accuracy. [ABSTRACT FROM AUTHOR]

Bexarotene (BEX) was approved by the FDA in 1999 for the treatment of cutaneous T-cell lymphoma (CTCL). The poor aqueous solubility causes the low bioavailability of the drug and thereby limits the clinical application. In this study, we developed a GCN-based deep learning model (CocrystalGCN) for in-silico screening of the cocrystals of BEX. The results show that our model obtained high performance relative to baseline models. The top 30 of 109 coformer candidates were scored by CocrystalGCN and then validated experimentally. Finally, cocrystals of BEX-pyrazine, BEX-2,5-dimethylpyrazine, BEX-methyl isonicotinate, and BEX-ethyl isonicotinate were successfully obtained. The crystal structures were determined by single-crystal X-ray diffraction. Powder X-ray diffraction, differential scanning calorimetry, and thermogravimetric analysis were utilized to characterize these multi-component forms. All cocrystals present superior solubility and dissolution over the parent drug. The pharmacokinetic studies show that the plasma exposures (AUC0−8h) of BEX-pyrazine and BEX-2,5-dimethylpyrazine are 1.7 and 1.8 times that of the commercially available BEX powder, respectively. This work sets a good example for integrating virtual prediction and experimental screening to discover the new cocrystals of water-insoluble drugs. [ABSTRACT FROM AUTHOR]

In recent years, with the rapid development of the Internet and information technology, video websites, shopping websites, and other portals have grown rapidly. However, malicious webpages can disguise themselves as benign websites and steal users' private information, which seriously threatens network security. Current detection methods for malicious webpages do not fully utilize the syntactic and semantic information in the web source code. In this paper, we propose a GCN-based malicious webpage detection method (GMWD), which constructs a text graph to describe and then a GCN model to learn the syntactic and semantic correlations within and between webpage source codes. We replace word nodes in the text graph with phrase nodes to better maintain the syntactic and semantic integrity of the webpage source code. In addition, we use the URL links appearing in the source code as auxiliary detection information to further improve the detection accuracy. The experiments showed that the proposed method can achieve 99.86% accuracy and a 0.137% false negative rate, achieving a better performance than other related malicious webpage detection methods. [ABSTRACT FROM AUTHOR]

Sign language recognition and translation can address the communication problem between hearing-impaired and general population, and can break the sign language boundariesy between different countries and different languages. Traditional sign language recognition and translation algorithms use Convolutional Neural Networks (CNNs) to extract spatial features and Recurrent Neural Networks (RNNs) to extract temporal features. However, these methods cannot model the complex spatiotemporal features of sign language. Moreover, RNN and its variant algorithms find it difficult to learn long-term dependencies. This paper proposes a novel and effective network based on Transformer and Graph Convolutional Network (GCN), which can be divided into three parts: a multi-view spatiotemporal embedding network (MSTEN), a continuous sign language recognition network (CSLRN), and a sign language translation network (SLTN). MSTEN can extract the spatiotemporal features of RGB data and skeleton data. CSLRN can recognize sign language glosses and obtain intermediate features from multi-view input sign data. SLTN can translate intermediate features into spoken sentences. The entire network was designed as end-to-end. Our method was tested on three public sign language datasets (SLR-100, RWTH, and CSL-daily) and the results demonstrated that our method achieved excellent performance on these datasets. [ABSTRACT FROM AUTHOR]

Link prediction is one of the most important methods to uncover evolving mechanisms of dynamic complex networks. Determining these links raises well-known technical challenges in terms of weak correlation, uncertainty and non-stationary. In this paper, we presented a novel gated graph convolutional network (GCN) based on spatio-temporal semi-variogram (STEM-GCN). It learns spacial and temporal features in order to achieve link prediction in the dynamic networks. In this STEM-GCN model, we first utilized the spatio-temporal semi-variogram to obtain the spacial and temporal correlations from the dynamic networks. Its spacial correlation helped us determine the hyper-parameters of STEM-GCN and speed up its training. The correlation smoothing strategy is also introduced to eliminate the noise through temporal correlation and to improve the accuracy of link prediction. Finally, the network dynamics are captured by propagating the spacial and temporal features between consecutive time steps with stacked memory cell structures. The extensive experiments on real data sets demonstrated the effectiveness of the proposed approach for link prediction in dynamic complex networks. [ABSTRACT FROM AUTHOR]

As the education of students attracts more and more attention, the task of graduation development prediction has gradually become a hot topic in academia and industry. The task of graduation development prediction aims to predict the employment category of students in advance via academic achievement data, which can help administrators understand students' learning status and set up a reasonable learning plan. However, existing research ignores the potential impact of social relationships on students' graduation development choices. To fully explore social relationships among students, we propose a Social-path Embedding-based Transformer Neural Network (SPE-TNN) for the task of graduation development prediction in this paper. Specifically, SPE-TNN is divided into the Social-path selection layer, the Social-path embedding layer, the Transformer layer, and the Multi-layer projection layer. Firstly, the Social-path selection layer is designed to find social relationships that impact graduation development and embed them into the student's performance features through the Social-path embedding layer. Secondly, the Transformer layer is adopted to balance the weights of the students' features. Finally, the Multi-layer projection layer is used to achieve the student graduation development prediction. Experimental results on the real-world datasets show that SPE-TNN outperforms the existing popular approaches. [ABSTRACT FROM AUTHOR]

In recent years, great progress has been made in the recognition of skeletal behaviors based on graph convolutional networks (GCNs). In most existing methods, however, the fixed adjacency matrix and fixed graph structure are used for skeleton data feature extraction in the spatial dimension, which usually leads to weak spatial modeling ability, unsatisfactory generalization performance, and an excessive number of model parameters. Most of these methods follow the ST-GCN approach in the temporal dimension, which inevitably leads to a number of non-key frames, increasing the cost of feature extraction and causing the model to be slower in terms of feature extraction and the required computational burden. In this paper, a gated temporally and spatially adaptive graph convolutional network is proposed. On the one hand, a learnable parameter matrix which can adaptively learn the key information of the skeleton data in spatial dimension is added to the graph convolution layer, improving the feature extraction and generalizability of the model and reducing the number of parameters. On the other hand, a gated unit is added to the temporal feature extraction module to alleviate interference from non-critical frames and reduce computational complexity. A channel attention mechanism based on an SE module and a frame attention mechanism are used to enhance the model's feature extraction ability. To prevent model degradation and ensure more stable training, residual links are added to each feature extraction module. The proposed approach was ultimately able to achieve 0.63% higher accuracy on the X-Sub benchmark with 4.46 M fewer parameters than GAT, one of the best SOTA methods. Inference speed of our model reaches as fast as 86.23 sequences/(second × GPU). Extensive experimental results further validate the effectiveness of our proposed approach on three large-scale datasets, namely, NTU RGB+D 60, NTU RGB+D 120, and Kinetics Skeleton. [ABSTRACT FROM AUTHOR]

Coal mining work has always been a high-risk job, although mining technology is now regularly very mature, many accidents still occur every year in various countries around the world, most of which are due to gas explosions, poisoning, asphyxiation and other accidents. Therefore it is important to monitor and predict both underground mine air quality. In this paper, we use the GCN spatio-temporal graph convolution method based on spectral domain for multivariate time series prediction of underground mine air environment. The correlation of these sequences is learned by a self-attentive mechanism, without a priori graph, and the adjacency matrix with an attention mechanism is created dynamically. The temporal and spatial features are learned by graph Fourier transform and inverse Fourier transform in TC module (temporal convolution) and GC module (graph convolution), respectively. Besides, the corresponding experimental predictions are performed on other public datasets. And a new loss function is designed based on the idea of residuals, which greatly improves the prediction accuracy. In addition, the corresponding experimental predictions were performed on other public datasets. The results show that this model has outstanding prediction ability and high prediction accuracy on most time-series prediction data sets. Through experimental verification, this model has high prediction accuracy for dealing with multivariate time series prediction problems, both for long-term and short-term prediction. [ABSTRACT FROM AUTHOR]

Featured Application: Blockchain-powered IoT Applications, such as agricultural applications, logistics, and smart factory data management. Blockchain network architecture is a promising technology for constructing highly secure Internet of Things (IoT) networks. IoT networks typically comprise various sensors and actuators. Blockchain network technology can be applied to secure control robots in smart factories or reliable drone deliveries in smart cities. The wide spread of transactions and shared smart contracts across blockchain networks guarantees ultimate network security. A typical wired blockchain network maintains sufficient redundancy within a stable configuration. However, IoT blockchain networks exhibit unavoidable instability. The dynamic configuration changes caused by flexible node membership make it impossible to achieve the same level of redundancy as a stable network. A trustworthy transaction spreading method provides practical transaction sharing for dynamic IoT networks. We propose a Q-learning framework and a graph convergence network (GCN) to search for the proper spreading path of each transaction. The proposed Q-learning framework determines the next spreading hop using node features. The GCN determines the reliable area based on the Q-learning results. The discovered reliable area guides the proper spreading path of transactions to the destination node. In addition, the proposed trustworthy transaction spreading was implemented over an InterPlanetary File system (IPFS). The IPFS-powered experiments confirmed the practicability of the proposed transaction spreading mechanism. [ABSTRACT FROM AUTHOR]

Dense and accurate 3D shape acquisition of objects by active-stereo technique has been an important research topic and intensively researched. One of the promising fields for active-stereo techniques is medical applications, such as 3D endoscope systems. In such systems, since a sensor is dynamically moved during the operation, single-frame shape reconstruction, a.k.a. oneshot scan, is necessary. For oneshot scan, there are several open problems, such as low resolution because of spatial coding, and unstable correspondence estimation between the detected patterns and the projected pattern because of irregular reflection. In this paper, we propose a solution for those problems. To increase the resolution, an accurate and stable interpolation method based on deep neural networks (DNNs) is proposed. Since most patterns used for oneshot scan are periodic, pixel-wise phase estimation can be achieved by detecting repetition in the pattern. A graph convolutional network (GCN), which is a deep neural network for graphs, is used for the correspondence problem. In the experiment, pixel-wise shape reconstruction results, as well as robust correspondence estimation using DNNs and a GCN, are shown. In addition, the effectiveness of the techniques is confirmed by comparing the proposed method with existing methods. [ABSTRACT FROM AUTHOR]

Graph convolutional networks (GCNs) based on convolutional operations have been developed recently to extract high-level representations from graph data. They have shown advantages in many critical applications, such as recommendation system, natural language processing, and prediction of chemical reactivity. The problem for the GCN is that its target applications generally pose stringent constraints on latency and energy efficiency. Several studies have demonstrated that field programmable gate array (FPGA)-based GCNs accelerators, which balance high performance and low power consumption, can continue to achieve orders-of-magnitude improvements in the inference of GCNs models. However, there still are many challenges in customizing FPGA-based accelerators for GCNs. It is necessary to sort out the current solutions to these challenges for further research. For this purpose, we first summarize the four challenges in FPGA-based GCNs accelerators. Then we introduce the process of the typical GNN algorithm and several examples of representative GCNs. Next, we review the FPGA-based GCNs accelerators in recent years and introduce their design details according to different challenges. Moreover, we compare the key metrics of these accelerators, including resource utilization, performance, and power consumption. Finally, we anticipate the future challenges and directions for FPGA-based GCNs accelerators: algorithm and hardware co-design, efficient task scheduling, higher generality, and faster development.

Previous video action recognition mainly focuses on extracting spatial and temporal features from videos or capturing physical dependencies among joints. The relation between joints is often ignored. Modeling the relation between joints is important for action recognition. Aiming at learning discriminative relation between joints, this paper proposes a joint spatial-temporal reasoning (JSTR) framework to recognize action from videos. For the spatial representation, a joints spatial relation graph is built to capture position relations between joints. For the temporal representation, temporal information of body joints is modeled by the intra-joint temporal relation graph. The spatial reasoning feature and the temporal reasoning feature are fused to recognize action from videos. The effectiveness of our method is demonstrated in three real-world video action recognition datasets. The experiment results display good performance across all of these datasets.

EEG-based human identification has gained a wide range of attention due to the further increase in demand for security. How to improve the accuracy of the human identification system is an issue worthy of attention. Using more features in the human identification system is a potential solution. However, too many features may cause overfitting, resulting in the decline of system accuracy. In this work, the graph convolutional neural network (GCN) was adopted for classification. Multiple features were combined and utilized as the structure matrix of the GCN. Because of the constant signal matrix, the training parameters would not increase as the structure matrix grows. We evaluated the classification accuracy on a classic public dataset. The results showed that utilizing multiple features of functional connectivity (FC) can improve the accuracy of the identity authentication system, the best results of which are at 98.56%. In addition, our methods showed less sensitivity to channel reduction. The method proposed in this paper combines different FCs and reaches high classification accuracy for unpreprocessed data, which inspires reducing the system cost in the actual human identification system. [ABSTRACT FROM AUTHOR]

This study alms at the automatic understanding of pedestrians' car-hailing intention in traffic scenes. Traffic scenes are highly complex, with a completely random spatial distribution of pedestrians. Different pedestrians use different behavior to express car-hailing intention, making it difficult to accurately understand the intention of pedestrians for autonomous taxis in complex scenes. A novel intention recognition algorithm with interpretability is proposed in this paper to solve the above problems. Firstly, we employ OpenPose to obtain skeleton data and the facial region. Then, we input the facial region into a facial attention network to extract the facial attention features and infer whether the pedestrian is paying attention to the ego-vehicle. In addition, the skeleton data are also input into a random forest classifier and GCN to extract both explicit and implicit pose features. Finally, an interpretable fusion rule is proposed to fuse the facial and pose features. The fusion algorithm can accurately and stably infer the pedestrians' intention and identify pedestrians with car-hailing intentions. In order to evaluate the performance of the proposed method, we collected road videos using experimental cars to obtain suitable datasets, and established the corresponding evaluation benchmarks. The experimental results demonstrate that the proposed algorithm has high accuracy and robustness. [ABSTRACT FROM AUTHOR]

Skeleton-based action recognition is a research hotspot in the field of computer vision. Currently, the mainstream method is based on Graph Convolutional Networks (GCNs). Although there are many advantages of GCNs, GCNs mainly rely on graph topologies to draw dependencies between the joints, which are limited in capturing long-distance dependencies. Meanwhile, Transformer-based methods have been applied to skeleton-based action recognition because they effectively capture long-distance dependencies. However, existing Transformer-based methods lose the inherent connection information of human skeleton joints because they do not yet focus on initial graph structure information. This paper aims to improve the accuracy of skeleton-based action recognition. Therefore, a Graph Skeleton Transformer network (GSTN) for action recognition is proposed, which is based on Transformer architecture to extract global features, while using undirected graph information represented by the symmetric matrix to extract local features. Two encodings are utilized in feature processing to improve joints' semantic and centrality features. In the process of multi-stream fusion strategies, a grid-search-based method is used to assign weights to each input stream to optimize the fusion results. We tested our method using three action recognition datasets: NTU RGB+D 60, NTU RGB+D 120, and NW-UCLA. The experimental results show that our model's accuracy is comparable to state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

At present, there is no uniform definition of annotation schemes for causal extraction, and existing methods are limited by the dependence of relations on long spans, which makes complex sentences such as multi-causal relations and nested causal relations difficult to extract. To solve these problems, a head-to-tail entity annotation method is proposed, which can express the complete semantics of complex causal relations and clearly describe the boundaries of entities. Based on this, a causal model, RPA-GCN (relation position and attention-graph convolutional networks), is constructed, incorporating GAT (graph attention network) and entity location perception. The attention layer is combined with a dependency tree to enhance the model's ability to perceive relational features, and a bi-directional graph convolutional network is constructed to further capture the deep interaction information between entities and relationships. Finally, the classifier iteratively predicts the relationship of each word pair in the sentence and analyzes all causal pairs in the sentence by a scoring function. Experiments on SemEval 2010 task 8 and the Altlex dataset show that our proposed method has significant advantages in solving complex causal extraction compared to state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Previous AutoML pruning works utilized individual layer features to automatically prune filters. We analyze the correlation for two layers from the different blocks which have a short-cut structure. It shows that, in one block, the deeper layer has many redundant filters which can be represented by filters in the former layer. So, it is necessary to take information from other layers into consideration in pruning. In this paper, a novel pruning method, named GraphPruning, is proposed. Any series of the network is viewed as a graph. To automatically aggregate neighboring features for each node, a graph aggregator based on graph convolution networks (GCN) is designed. In the training stage, a PruningNet that is given aggregated node features generates reasonable weights for any size of the sub-network. Subsequently, the best configuration of the Pruned Network is searched by reinforcement learning. Different from previous work, we take the node features from a well-trained graph aggregator instead of the hand-craft features, as the states in reinforcement learning. Compared with other AutoML pruning works, our method has achieved the state-of-the-art under the same conditions on ImageNet-2012. [ABSTRACT FROM AUTHOR]

Epilepsy is a chronic brain disease that causes persistent and severe damage to the physical and mental health of patients. Daily effective prediction of epileptic seizures is crucial for epilepsy patients especially those with refractory epilepsy. At present, a large number of deep learning algorithms such as Convolutional Neural Networks and Recurrent Neural Networks have been used to predict epileptic seizures and have obtained better performance than traditional machine learning methods. However, these methods usually transform the Electroencephalogram (EEG) signal into a Euclidean grid structure. The conversion suffers from loss of adjacent spatial information, which results in deep learning models requiring more storage and computational consumption in the process of information fusion after information extraction. This study proposes a general Graph Convolutional Networks (GCN) model architecture for predicting seizures to solve the problem of oversized seizure prediction models based on exploring the graph structure of EEG signals. As a graph classification task, the network architecture includes graph convolution layers that extract node features with one-hop neighbors, pooling layers that summarize abstract node features; and fully connected layers that implement classification, resulting in superior prediction performance and smaller network size. The experiment shows that the model has an average sensitivity of 96.51%, an average AUC of 0.92, and a model size of 15.5 k on 18 patients in the CHES-MIT scalp EEG dataset. Compared with traditional deep learning methods, which require a large number of parameters and computational effort and are demanding in terms of storage space and energy consumption, this method is more suitable for implementation on compact, low-power wearable devices as a standard process for building a generic low-consumption graph network model on similar biomedical signals. Furthermore, the edge features of graphs can be used to make a preliminary determination of locations and types of discharge, making it more clinically interpretable. [ABSTRACT FROM AUTHOR]

Mobile Laser Scanning (MLS) system can provide high-density and accurate 3D point clouds that enable rapid pavement crack detection for road maintenance tasks. Supervised learning-based algorithms have been proved pretty effective for handling such a large amount of inhomogeneous and unstructured point clouds. However, these algorithms often rely on a lot of annotated data, which is labor-intensive and time-consuming. This paper presents a semi-supervised point-level approach to overcome this challenge. We propose a graph-widen module to construct a reasonable graph structure for point clouds, increasing the detection performance of graph convolutional networks (GCN). The constructed graph characterizes the local features from a small amount of annotated data, avoiding information loss and dramatically reduces the dependence on annotated data. The MLS point clouds acquired by a commercial RIEGL VMX-450 system are used in this study. The experimental results demonstrate that our method outperforms the state-of-the-art point-level methods in terms of recall, F1 score, and efficiency while achieving comparable accuracy. [ABSTRACT FROM AUTHOR]