No abstract

No abstract

Image classification is an important application of deep learning. In a typical classification task, the classification accuracy is strongly related to the features that are extracted via deep learning methods. An autoencoder is a special type of neural network, often used for dimensionality reduction and feature extraction. The proposed method is based on the traditional autoencoder, incorporating the "distance" information between samples from different categories. The model is called a semisupervised distance autoencoder. Each layer is first pre-trained in an unsupervised manner. In the subsequent supervised training, the optimized parameters are set as the initial values. To obtain more suitable features, we use a stacked model to replace the basic autoencoder structure with a single hidden layer. A series of experiments are carried out to test the performance of different models on several datasets, including the MNIST dataset, street view house numbers (SVHN) dataset, German traffic sign recognition benchmark (GTSRB), and CIFAR-10 dataset. The proposed semi-supervised distance autoencoder method is compared with the traditional autoencoder, sparse autoencoder, and supervised autoencoder. Experimental results verify the effectiveness of the proposed model.

In traffic systems, traffic forecasting is a critical issue, which has attracted much interest from researchers. It is a challenging task due to the complex spatial‐temporal patterns of traffic data. Previous works focus on designing complex graph‐based neural networks to model spatial‐temporal dependencies from data. By using graphs to represent road networks, these works capture spatial patterns with graph convolutions. However, graphs cannot fully represent spatial relations from road networks. It limits the performance of graph‐based methods. In this paper, we propose a directed hypergraph neural network architecture, Directed Hypergraph Attention Network(DHAT), for traffic forecasting. Unlike previous works, DHAT introduces a directed hypergraph to represent road networks. Compared with graphs, directed hypergraphs could represent spatial information from graphs and outperform them in modeling complex directed relations among multiple nodes. Based on the directed hypergraph, a directed hypergraph convolution is proposed to exploit spatial relations among traffic series. By combining the proposed convolution and attention mechanisms, DHAT can effectively achieve promising predictions for traffic forecasting. To evaluate the performance of DHAT, we have conducted extensive experiments on four real‐world traffic datasets. Compared with other baselines, experimental results show that DHAT reduces Mean Absolute Error by 0.03–0.64 on these datasets.

In real applications, label noise and feature noise are two main noise sources. Similar to feature noise, label noise imposes great detriment on training classification models. Motivated by successful application of deep learning method in normal classification problems, this paper proposes a new framework called LNC-SDAE to handle those datasets corrupted with label noise, or so-called inaccurate supervision problems. The LNC-SDAE framework contains a preliminary label noise cleansing part and a stacked denoising auto-encoder. In preliminary label noise cleansing part, the K-fold cross-validation thought is applied for detecting and relabeling those mislabeled samples. After being preprocessed by label noise cleansing part, the cleansed training dataset is then input into the stacked denoising auto-encoder to learn robust representation for classification. A corrupted UCI standard dataset and a corrupted real industrial dataset are used for test, both of which contain a certain proportion of label noise (the ratio changes from 0% to 30%). The experiment results prove the effectiveness of LNC-SDAE, the representation learnt by which is shown robust.