Convolutional neural network (CNN) is quite popular in computer vision, especially in image classification with excellent performance. However, limited by the convolution kernels, CNN-based classifiers are hard to extract global feature from the original image, while exact object locations in the environment are included in the global feature. One popular way to improve global feature extraction performance is to use graph neural network (GNN) which can aggregate global information through the connection relationship of different nodes. In this work, a novel end-to-end graph neural network architecture is proposed, in which local and global-attention feature are used simultaneously to achieve more accurate predictions. In this architecture, a CNN block is designed to learn local feature and graph convolutional neural network (GCN) is used to learn global feature. Global-attention feature for final prediction is down-sampled from global feature by the proposed global multi-head self-attention pooling (GMSAPool) based on self-attention mechanism, which reconstructs the input graph by introducing virtual node and automatically assigns different weights to each node to obtain a more representative global-attention feature. In addition, the proposed architecture can be trained without converting images to graphs in advance, and the computational burden can also be reduced. This approach is demonstrated on three open datasets (Agricultural Disease, Caltech256 and CIFAR-100) to validate the effectiveness. The determined experimental results showed that: (1) The proposed model achieve 84.46%, 77.80%, and 83.33% on the Macro-F1 in three datasets respectively, improving over the best baselines; (2) Global-attention feature that is more conducive to the final prediction is extracted by GMSAPool from numerous nodes, in which Macro-P,Macro-R and Macro-F1 are respectively improved 3.655%,1.12%,2.715% on average.
Current semantic segmentation algorithms are often burdened by high computational complexity and inadequate boundary localization accuracy in complex scenarios of industrial manufacturing. To achieve lightweight semantic segmentation with high accuracy, we propose a semantic segmentation algorithm named L-DeepLabV3+. First, a CB-MobileNetV3 is proposed as the backbone network, which reduces the computational burden. Second, we propose a DSCA_ASPP module to compensate for the loss of some detailed information in the encoder and enhance the perception of boundary information. Within the DSCA_ASPP module, we propose dilated separable convolution to address the issue of dilated convolutional meshing. Finally, we introduce the Haar Wavelet Transform with Deepwise Separable Convolution (HWTS) module as a replacement for the 3×3 convolution in the decoder. It consists of a deepwise separable convolution and Haar wavelet transform module, which further reduces the number of model parameters and improves the feature representation in the decoder. Experimental results demonstrate that L-DeepLabV3+ achieves an 82.7% mIoU on the PASCAL VOC2012 dataset, utilizing only 2.73 million parameters and a floating-point computation amounting to 43.015 G. The L-DeepLabV3+ model has ∼4% of the number of parameters compared to the traditional DeepLabV3+, resulting in a significant reduction in computational burden. Furthermore, L-DeepLabV3+ outperforms common semantic segmentation algorithms while using fewer model parameters and enhancing segmentation performance. Moreover, L-DeepLabV3+ exhibits excellent performance in detecting defects in industrial ceramic substrates.
The users on the Internet have been growing exponentially, and tag recommendation can automatically provide users with a selection of tags of interest to meet their personalized needs. Users can utilize these tags to freely annotate their favorite resources, making them efficient and fast in retrieving related resources. Tensor factorization methods are commonly used in tag recommendation at present. These methods model the user × item × tag interactions, transform the latent feature representations of users, items and tags into low-rank matrices and use inner products for prediction. However, the problem of using inner product is that it does not satisfy the triangle inequality, it ignores the distance relationship among entity pairs and cannot capture the fine-grained preference information. Metric learning in recommendation domains focus on using pairwise loss, which assumes that different categories (such as users, items and tags) have fixed margins. Different categories often have different intra-class variations. With fixed margins, it is often difficult to accurately distinguish between positive and negative samples, thus reducing recommendation performance and limiting the expression ability of the model. In this study, the metric learning method is used to explore the distance relationship among user × item × tag triplet, and the existing metric learning based methods (namely LRML, CML, SML) are applied to the tag recommendation. A pairwise metric learning method with angular margin is proposed, named PMLT. The pairwise distance relationship between user-tag and item-tag is modeled for the information of different entities. And an extra angular margin regularizer is added to the original pairwise loss to control the size of angular margin for user-tag and item-tag respectively. The strength of the constrained angular margin regularizer is controlled to dynamically adjust the distance changes of entity. This method constrains the fixed margin and also the angular margin of user-tag and item-tag. Compared with the traditional metric learning method, this method can capture additional relationship structure and has good recommendation performance. Finally, we conducted extensive experiments on two datasets, LastFm and Movielens, and the experimental results showed that the proposed method PMLT outperform the state-of-the-art baseline in the evaluation metrics Recall@N and NDCG@N, and obtain better prediction quality. We also analyze the influence of different parameters and internal components on the performance of the proposed method, which improves the interpretability of the proposed method.
With the rapid development of artificial intelligence technology, commercial robots have gradually entered our daily lives. In order to promote product dissemination, shopping guide robots are a new service options of commerce platforms that use tag recommendation systems to identify users' intentions. A large number of applications combine user historical tagging information with the multi‐round dialogue ability of shopping guide robots to help users efficiently search for and retrieve products of interest. Recently, tensor decomposition methods have become a common approach for modelling entity interaction relationships in tag recommendation systems. However, due to the sparsity of data, these methods only consider low‐order information of entities, making it difficult to capture the higher‐order collaborative signals among entities. Recommendation methods by autoencoders can effectively extract abstract feature representations while they only focus on the two‐dimensional relationship between users and items, ignoring the interaction relationship among users, items and tags in real complex recommendation scenarios. The authors focus on modelling the similarity relationship among entities and propose a method called deep feature fusion tag (DFFT) based on the deep feature fusion of stacked denoising autoencoders. This method can extract high‐order information with different embedding dimensions and fuse them in a unified framework. To extract robust feature representations, the authors inject random noise (mask‐out/drop‐out noise) into the tag information corresponding to users and items to generate corrupted input data, and then utilise autoencoders to encode the interaction relationship among entities. To further obtain the interaction relationship with different dimensions, different encoding layers are stacked and combined to produce a better expanded model which can reinforce each other. Finally, a decoding component is used to reconstruct the original input data. According to the experimental results on two common datasets, the proposed DFFT method outperforms other baselines in terms of the F1@N, NDCG@N and Recall@N evaluation metrics.
1 October 2015 in Chinese Journal of Chemical Engineering