Capsule network is a promising model in computer vision. It has achieved excellent results on simple datasets such as MNIST, but the performance deteriorates as data becomes complicated. In order to address this issue, we propose a deep capsule network in this paper. To deepen the capsule network, we present a new tensor capsule based routing algorithm and the corresponding convolution operation. Compared to vector capsules, tensor capsules can capture more instance-level information. Together, the relevant convolution operation is beneficial for reducing the amount of parameters in the routing process. Furthermore, we propose a dropout mechanism for vectors and tensors in order to alleviate the potential overfitting problem. Finally, we also inject the multi-scale capsules of the middle layers into a multi-scale decoder to pursue more details of an image and more clear reconstructed image. Experimental results on CIFAR10, Fashion-MNIST, and SVHN demonstrate that the proposed deep tensor network can achieve very competitive performance compared to other state-of-the-art capsule networks.
3 January 2023 in Environmental Science and Pollution Research
Energy consumption of buildings is correlated with environmental factors such as weather and temperature. Developing a reliable energy forecasting model requires a large-scale, high-quality dataset. However, privacy regulations mandate that data be stored locally, prohibiting data sharing. As a solution, a Federated Building Energy Forecasting framework has emerged, supporting machine learning using decentralized energy data. This framework facilitates the communication of deep learning models between a server and multiple clients. However, it remains vulnerable to server crashes, which can disrupt the entire network’s functionality. To address this issue, multi-server federated learning strategies have been proposed to reduce dependency on a single server and enhance reliability. Additionally, the meteorological variable of wind speed has been integrated to improve energy forecasting through multi-task learning. A multi-task learning method is proposed to forecast energy load and wind speed simultaneously, employing a mixture-of-experts (MoE) architecture. Furthermore, a new metaformer neural network has been designed using the Taylor expansion approximation to simplify self-attention mechanisms. Extensive experiments have evaluated the effectiveness of multi-task learning and multi-server federated training. The results demonstrate that forecasting accuracies have improved by 10% to 30% compared to existing state-of-the-art transformer forecasting models. Additionally, incorporating wind speed within the MoE framework benefits energy forecasting. Multi-server strategies also help mitigate data scarcity issues through strategic random clustering.
With the rapid growth of advanced driving assistance technology, how to implement humancar interaction humanely and personally becomes increasingly important. Existing methods typically rely on hand-crafted data to conduct training, which cannot fully extract information hidden in text. In this paper, we tackle the problem of label data acquisition difficulties and inadequate feature extraction via semisupervised learning and multi-granularity learning. To fulfill the goal of label data acquisition and feature extraction, we present a novel semi-supervised multi-granularity convolutional neural networks (CNNs)- based (SSMGCNNs-based) model for an application in human-car interaction, which consists of the twoview- embedding (TVE) module and the multi-granularity CNNs (MGCNNs) module. The TVE module learns embeddings of text regions from the unlabeled user command data set and then integrate the learned tv-embeddings into MGCNNs, so that the learned tv-embedded regions are used as an additional input into MGCNNs’ convolution layer to solve the problem of data annotation. MGCNNs can fully extract information hidden in text by multiple convolution kernels of the same convolution layer. We compared our model with some state-of-the-art machine learning models. On the car operation command data set, the simulation results demonstrated that, compared with CNNs, our method respectively improved 5.13%, 5.64%, 3.60% and 5.34% on precision, recall, F-1 and training loss.
Person re-identification (Re-ID) aims at retrieving a person's identification across multiple non-overlapping cameras. While recent Re-ID methods have achieved significant success on a number of benchmark datasets, most of them are still insufficient in the scenes with unified or highly similar dressing like construction sites, schools and factories. To address this problem, we propose a supervisory mask attention network (SMA-Net). Our approach combines two key components: (1) ROI mask mapping (RMM) is a supervisory branch to provide ROI mappings that divide a person region into several parts; (2) Partial mask attention (PMA) integrates channel and space attention mechanisms that focus on local features and different accessories in each ROI. Therefore, the network can pay more attention to the local features and different accessories herein. Compared with the state of the art methods, SMA-Net demonstrates excellent performance on our dataset of construction scenes, with improvement of 6.65% in mAP and 3.8% in Rank-1 accuracy.