As an image enhancement technology, multi-modal fusion primarily aims to retain salient information from multi-source pairs in a single image, generating imaging that contains complementary features and can facilitate downstream visual tasks. However, dual-stream methods with convolutional neural networks (CNNs) as backbone predominantly have limited receptive fields, whereas Transformers are time-consuming, both lack the exploration of cross-domain information. This study proposes...

In recent years, distantly-supervised relation extraction has achieved a certain success by using deep neural networks. Distant Supervision (DS) can automatically generate large-scale annotated data aligning entity pairs from Knowledge Bases (KB) to sentences. However, these DS-generated datasets inevitably have wrong labels that result in incorrect evaluation scores during testing, which may mislead the researchers. To solve this problem, we build new dataset NYTH, where use as training and...

Research on traffic data analysis is becoming more available and important. One of the key challenges how to accurately decompose high-dimensional, noisy observation flow matrix into sub-matrices that correspond different classes which builds a foundation for prediction, abnormal detection missing imputation. While in traditional research, Principal Component Analysis (PCA) usually used analysis. However, corrupted by large volume anomalies, resulting principal components will be...

Traffic flow prediction is a fundamental component in Intelligent Transportation Systems (ITS). Nearest neighbor based nonparametric regression method classic data-driven for traffic prediction. Modern data collection technologies provide the opportunity to represent various features of nonlinear complex system which also bring challenges fuse multiple sources data. Firstly, Euclidean distance metric models that treat each feature with equal weight not effective multi-source high-dimension...

Legal judgment prediction (LJP) is used to predict results based on the description of individual legal cases. In order be more suitable for actual application scenarios in which case has cited multiple articles and charges, we formulate as a label learning problem present deep model that can effectively encode content each via multi-residual convolution neural network semantics law an article encoder. An article-wise attention mechanism proposed fuse two types encoded information....

Abstract We report on a grating-free fiber chirped pulse amplifier (CPA) at 2.8 μm for the first time. The CPA system adopted Er:ZBLAN with large anomalous dispersion as stretcher and germanium (Ge) rods compressor compact structure. High-energy picosecond pulses of 2.07 μJ were generated repetition rate 100 kHz. Using highly dispersive Ge rods, amplified compressed to 408 fs energy 0.57 μJ, resulting in peak power approximately 1.4 MW. A spectral broadening phenomenon main was observed,...

Nonparametric regression is a classic method for short-term traffic flow forecasting in Intelligent Transportation Systems (ITS). Feature space construction and distance metric selection are two important parts nonparametric regression. Few of previous works have taken both these aspects into account together. In addition, how to use information related stations network scale key improve the performance ITS. this paper, we propose novel three-stage framework based on KNN handle issues above...

Learning temporal causal structures between time series is one of key tools for analyzing data. Most previous works focuse on learning with static relationships. However, in many real world applications, such as climate environment and transportation system, the vary dramatically over time. In this paper, we propose a probabilistic dynamic (PDC) model based Lasso-Granger to uncover dependencies. Specifically, PDC infers different state varying data each unified model. We devise...

We demonstrated the generation of a nearly diffraction-limited picosecond pulse from large-mode-area (LMA) fluoride fiber amplifier. Seeded with mode-locked oscillator at 2.8 µm, LMA Er:ZBLAN amplifier delivered 16 µJ duration 70 ps 5 kHz. The beam was obtained 50 µm using fundamental mode excitation technique, measured M

With the rapid development of Internet, security network multimedia data has attracted increasingly more attention. The moving target defense (MTD) and cyber mimic (CMD) approaches provide a new way to solve this problem. To enhance data, paper proposes encryption box for security. can directly access where device is located, automatically complete negotiation, safe convenient services, effectively prevent attacks. According principles dynamization, diversification, randomization, uses...

Xiabing Zhou, Zhongqing Wang, Shoushan Li, Guodong Min Zhang. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.

With the development of cloud computing, high-capacity reversible data hiding in an encrypted image (RDHEI) has attracted increasing attention. The main idea RDHEI is that owner encrypts a cover image, and then hider embeds secret information image. key, receiver can extract embedded from hidden image; with encryption reconstructs original In this paper, we embed form random bits or scanned documents. proposed method takes full advantage spatial correlation images to vacate room for...

Purpose The railway signal equipment failure diagnosis is a vital element to keep the system operating safely. One of most difficulties in uncertainty causality between consequence and cause for accident. traditional method solve this problem based on Bayesian Network, which needs rigid independent assumption basis prior probability knowledge but ignoring semantic relationship analysis. This paper aims perform through new way that emphasis mining relationships. Design/methodology/approach...

Learning with incomplete data remains challenging in many real-world applications especially when the is high-dimensional and dynamic. Many imputation-based algorithms have been proposed to handle data, where these use statistics of historical information remedy missing parts. However, methods merely structural existing which are very helpful for sharing between complete entries ones. For example, traffic system, some group temporal smoothness exist structure. In this paper, we propose...

Legal Judgment Prediction aims to automatically predict judgment outcomes based on descriptions of legal cases and established law articles, has received increasing attention. In the preliminary work, several problems still have not been adequately solved. One is how utilize limited but valuable label information. Existing methods mostly ignore gap between description articles cases, directly integrate them. Second, most studies mutual constraint among subtasks, such as logically or...

Existing approaches to Emotion Recognition in Conversation (ERC) use a fixed context window recognize speakers' emotion, which may lead either scantiness of key or interference redundant context. In response, we explore the benefits variable-length and propose more effective approach ERC. our approach, leverage different windows when predicting emotion utterances. New modules are included realize context: 1) two speaker-aware units, explicitly model inner- inter-speaker dependencies form...

Nearest neighbor based nonparametric regression is a classic data-driven method for traffic flow prediction in intelligent transportation systems (ITS). Performances of those models depend heavily on the similarity or distance metric used to search nearest neighborhood. Metric learning algorithms have been developed learn metrics from data recent years. In real-world application, multiple forecasting tasks are set since there lots road sections and detector points network. Previous works...

Emotion Recognition in Conversation (ERC) aims to recognize the emotion for each utterance a conversation automatically. Due difficulty of collecting and labeling, this task lacks dataset corpora available on large scale. This increases finishing supervised training required by large-scale neural networks. Introducing generative conversational can assist with modeling dialogue. However, spatial distribution feature vectors source target domains is inconsistent after introducing external...