Convolutional neural networks (CNNs) utilize local translation invariance in the Euclidean domain and have remarkable achievements in computer vision tasks. However, there are many data types with non-Euclidean structures, such as social networks, chemical molecules, knowledge graphs, etc., which are crucial to real-world applications. The graph convolutional neural network (GCN), as a derivative of CNNs for non-Euclidean data, was established for non-Euclidean graph data. In this paper, we mainly survey the progress of GCNs and introduce in detail several basic models based on GCNs. First, we review the challenges in building GCNs, including large-scale graph data, directed graphs and multi-scale graph tasks. Also, we briefly discuss some applications of GCNs, including computer vision, transportation networks and other fields. Furthermore, we point out some open issues and highlight some future research trends for GCNs.

In the lightweight convolutional neural network model, the pointwise convolutional structure occupies most of the parameters and computation amount of the model. Therefore, improving the pointwise convolution structure is the best choice to optimize the lightweight model. Aiming at the problem that the pointwise convolution in MobileNetV1 and MobileNetV2 consumes too many computation resources, we designed the novel Ghost-PE and Ghost-PC blocks. First, in order to optimize the channel expanded pointwise convolution with the number of input channels less than the output, Ghost-PE makes full use of the feature maps generated by main convolution of the Ghost module, and adds global average pooling and depth convolution operation to enhance the information of feature maps generated through cheap convolution. Second, in order to optimize the channel compressed pointwise convolution with the number of input channels more than the output, Ghost-PC adjusts the Ghost-PE block to make full use of the features generated by cheap convolution to enhance the feature channel information. Finally, we optimized MobileNetV1 and MobileNetV2 models by Ghost-PC and Ghost-PE blocks, and then tested on Food-101, CIFAR and Mini-ImageNet datasets. Compared with other methods, the experimental results show that Ghost-PE and Ghost-PC still maintain a relatively high accuracy in the case of a small number of parameters.

Open-world semi-supervised learning (OWSSL) has received significant attention since it addresses the issue of unlabeled data containing classes not present in the labeled data. Unfortunately, existing OWSSL methods still rely on a large amount of labeled data from seen classes, overlooking the reality that a substantial amount of labels is difficult to obtain in real scenarios. In this paper, we explored a new setting called open-world barely-supervised learning (OWBSL), where only a single label was provided for each seen class, greatly reducing labeling costs. To tackle the OWBSL task, we proposed a novel framework that leveraged augmented pseudo-labels generated for the unlabeled data. Specifically, we first generated initial pseudo-labels for the unlabeled data using visual-language models. Subsequently, to ensure that the pseudo-labels remained reliable while being updated during model training, we enhanced them using predictions from weak data augmentation. This way, we obtained the augmented pseudo-labels. Additionally, to fully exploit the information from unlabeled data, we incorporated consistency regularization based on strong and weak augmentations into our framework. Our experimental results on multiple benchmark datasets demonstrated the effectiveness of our method.

In this paper, we introduced a novel group self-supervised learning approach designed to improve visual representation learning. This new method aimed to rectify the limitations observed in conventional self-supervised learning. Traditional methods tended to focus on embedding distortion-invariant in single-view features. However, our belief was that a better representation can be achieved by creating a group of features derived from multiple views. To expand the siamese self-supervised architecture, we increased the number of image instances in each crop, enabling us to obtain an average feature from a group of views to use as a distortion, invariant embedding. The training efficiency has greatly increased with rapid convergence. When combined with a robust linear protocol, this group self-supervised learning model achieved competitive results in CIFAR-10, CIFAR-100, Tiny ImageNet, and ImageNet-100 classification tasks. Most importantly, our model demonstrated significant convergence gains within just 30 epochs as opposed to the typical 1000 epochs required by most other self-supervised techniques.