Stacked Attention Networks for Referring Expressions Comprehension

Referring expressions comprehension is the task of locating the image region described by a natural language expression,which refer to the properties of the region or the relationships with other regions.Most previous work handles this problem by selecting the most relevant regions from a set of candidate regions,when there are many candidate regions in the set these methods are inefficient.Inspired by recent success of image captioning by using deep learning methods,in this paper we proposed a framework to understand the referring expressions by multiple steps of reasoning.We present a model for referring expressions comprehension by selecting the most relevant region directly from the image.The core of our model is a recurrent attention network which can be seen as an extension of Memory Network.The proposed model capable of improving the results by multiple computational hops.We evaluate the proposed model on two referring expression datasets:Visual Genome and Flickr30k Entities.The experimental results demonstrate that the proposed model outperform previous state-of-the-art methods both in accuracy and efficiency.We also conduct an ablation experiment to show that the performance of the model is not getting better with the increase of the attention layers.

Dynamic prediction of landslide displacement using singular spectrum analysis and stack long short-term memory network

An accurate landslide displacement prediction is an important part of landslide warning system. Aiming at the dynamic characteristics of landslide evolution and the shortcomings of traditional static prediction models, this paper proposes a dynamic prediction model of landslide displacement based on singular spectrum analysis(SSA) and stack long short-term memory(SLSTM) network. The SSA is used to decompose the landslide accumulated displacement time series data into trend term and periodic term displacement subsequences. A cubic polynomial function is used to predict the trend term displacement subsequence, and the SLSTM neural network is used to predict the periodic term displacement subsequence. At the same time, the Bayesian optimization algorithm is used to determine that the SLSTM network input sequence length is 12 and the number of hidden layer nodes is 18. The SLSTM network is updated by adding predicted values to the training set to achieve dynamic displacement prediction. Finally, the accumulated landslide displacement is obtained by superimposing the predicted value of each displacement subsequence. The proposed model was verified on the Xintan landslide in Hubei Province, China. The results show that when predicting the displacement of the periodic term, the SLSTM network has higher prediction accuracy than the support vector machine(SVM) and auto regressive integrated moving average(ARIMA). The mean relative error(MRE) is reduced by 4.099% and 3.548% respectively, while the root mean square error(RMSE) is reduced by 5.830 mm and 3.854 mm respectively. It is concluded that the SLSTM network model can better simulate the dynamic characteristics of landslides.

Prediction of Disease Transmission Risk in Universities Based on SEIR and Multi-hidden Layer Back-propagation Neural Network Model

Against the background of regular epidemic prevention and control,in order to ensure the return of teachers to work,students to return to school and safe operation of schools,the risk of disease transmission is analyzed in key areas such as university canoons,auditoriums,teaching buildings and dormitories.The risk model of epidemic transmission in key regions of universities is established based on the improved SEIR model,considering the four groups of people,namely susceptible,latent,infected and displaced,and their mutual transformation relationship.After feature post-processing,the selected feature parameters are processed with monotone non-decreasing and smoothing,and used as noise-free samples of stacked sparse denoising automatic coding network to train the network.Then,the feature vectors after dimensionality reduction of the stacked sparse denoising automatic coding network are used as the input of the multi-hidden layer back-propagation neural network,and these features are used as tags to carry out fitting training for the network.The results show that the implementation of control measures can reduce the number of contacts between infected people and susceptible people,reduce the transmission rate of single contact,and reduce the peak number of infected people and latent people by 61%and 72%respectively,effectively controlling the disease spread in key regions of universities.Our method is able to accurately predict the number of infections.

Multipath affinage stacked-hourglass networks for human pose estimation

Recently,stacked hourglass network has shown outstanding performance in human pose estimation.However,repeated bottom-up and top-down stride convolution operations in deep convolutional neural networks lead to a significant decrease in the initial image resolution.In order to address this problem,we propose to incorporate affinage module and residual attention module into stacked hourglass network for human pose estimation.This paper introduces a novel network architecture to replace the stacked hourglass network of up-sampling operation for getting high-resolution features.We refer to the architecture as an affinage module which is critical to improve the performance of the stacked hourglass network.Additionally,we also propose a novel residual attention module to increase the supervision of up-sample process.The effectiveness of the introduced module is evaluated on standard benchmarks.Various experimental results demonstrated that our method can achieve more accurate and more robust human pose estimation results in images with complex background.

NfvInsight:A Framework for Automatically Deploying and Benchmarking VNF Chains

With the advent of virtualization techniques and software-defined networking(SDN),network function virtualization(NFV)shifts network functions(NFs)from hardware implementations to software appliances,between which exists a performance gap.How to narrow the gap is an essential issue of current NFV research.However,the cumbersomeness of deployment,the water pipe effect of virtual network function(VNF)chains,and the complexity of the system software stack together make it tough to figure out the cause of low performance in the NFV system.To pinpoint the NFV system performance,we propose NfvInsight,a framework for automatic deployment and benchmarking VNF chains.Our framework tackles the challenges in NFV performance analysis.The framework components include chain graph generation,automatic deployment,and fine granularity measurement.The design and implementation of each component have their advantages.To the best of our knowledge,we make the first attempt to collect rules forming a knowledge base for generating reasonable chain graphs.NfvInsight deploys the generated chain graphs automatically,which frees the network operators from executing at least 391 lines of bash commands for a single test.To diagnose the performance bottleneck,NfvInsight collects metrics from multiple layers of the software stack.Specifically,we collect the network stack latency distribution ingeniously,introducing only less than 2.2%overhead.We showcase the convenience and usability of NfvInsight in finding bottlenecks for both VNF chains and the underlying system.Leveraging our framework,we find several design flaws of the network stack,which are unsuitable for packet forwarding inside one single server under the NFV circumstance.Our optimization for these flaws gains at most 3x performance improvement.