摘要:
Visual Dialog aims to answer an appropriate response based on a multi-round dialog history and a given image. Existing methods either focus on semantic interaction, or implicitly capture coarse-grained structural interaction (e.g., pronoun co-references). The fine-grained and explicit structural interaction feature for dialog history is seldom explored, resulting in insufficient feature learning and difficulty in capturing precise context. To address these issues, we propose a structure-aware dual-level graph interactive network (SDGIN) that integrates verb-specific semantic roles and co-reference resolution to explicitly capture context structural features for discriminative and generative tasks in visual dialog. Specifically, we create a novel structural interaction graph that injects syntactic knowledge priors into dialog by introducing semantic role labeling that imply which words are sentence stems. Furthermore, considering the single perspective limitation of previous algorithms, we design a dual-perspective mechanism that learns fine-grained token-level context structure features and coarse-grained utterance-level interactions in parallel. It possess an elegant view to explore precise context interactions, realizing the mutual complementation and enhancement of different granularity features. Experimental results show the superiority of our proposed approach. Compared to other task-specific models, our SDGIN outperforms previous models and achieves a significant improvement on the benchmark dataset VisDial v1.0.
作者机构:
[Muyang Mei; Wei Li; Zhongshuai Feng] Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and Technology, Wuhan 430074, China;[Yuan Li; Mengchao Niu; Jiaye Zhu] School of Computer Science, Central China Normal University, Wuhan 430079, China;[Ming Luo] State Key Laboratory of Optical Communication Technologies and Networks, China Information and Communication Technologies Group Corporation, Wuhan 430205, China;[Xuefeng Wu; Liang Mei] Fiberhome Telecommunication Technologies Co., Ltd., Wuhan 430205, China;[Qianggao Hu; Yi Jiang; Xuefeng Yang] Accelink Technologies Co., Ltd. Wuhan 430205, China
通讯机构:
[Wei Li] W;Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and Technology, Wuhan 430074, China
摘要:
We propose using physical-informed neural network (PINN) for power evolution prediction in bidirectional Raman amplified WDM systems with Rayleigh backscattering (RBS). Unlike models based on data-driven machine learning, PINN can be effectively trained without preparing a large amount of data in advance and can learn the potential rules of power evolution. Compared to previous applications of PINN in power prediction, our model considers bidirectional Raman pumping and RBS, which is more practical. We experimentally demonstrate power evolution prediction of 200 km bidirectional Raman amplified wavelength-division multiplexed (WDM) system with 47 channels and 8 pumps using PINN. The maximum prediction error of PINN compared to experimental results is only 0.38 dB, demonstrating great potential for application in power evolution prediction. The power evolution predicted by PINN shows good agreement with the results simulated by traditional numerical method, but its efficiency is more suitable for establishing models and calculating noise, providing convenience for subsequent power configuration optimization.
摘要:
A detailed theoretical study is conducted on the nonlinear interference in the same-wavelength bidirectional coherent optical fiber communication systems. The Gaussian noise (GN) model used to evaluate nonlinear interference (NLI) in unidirectional systems is applied and extended to bidirectional transmission scenarios. The extended NLI model shows that in a bidirectional transmission communication system, the backward signal almost does not introduce additional nonlinear crosstalk to the forward signal due to the strong walk-off effect between forward and backward transmitted signals. Specifically, the ratio of the nonlinear crosstalk introduced by the forward and backward signals is about 21 dB, which means that the traditional GN model is also applicable in the bidirectional scenario. This conclusion is validated on the platform of a same-wavelength bidirectional coherent optical communication system based on Optisystem software.
摘要:
A direction-aware augmented spatial keyword top- $k$ query (DAT $k\text{Q}$ ) returns the top- $k$ objects based on a ranking function that considers spatial distance, textual similarity, query numeric attributes, and query direction. When a user initiates a DAT $k\text{Q}$ , some user-desired objects (missing objects) may not appear in the query result set, and then the user wonders why they do not appear, which is called the why-not question. This paper focuses on answering why-not questions on DAT $k$ Qs. We first discuss how to obtain the refined query direction by analyzing the position relationship between missing objects and original query direction in Polar coordinates. Then a DAPC index structure is designed, which can cut down irrelevant search space based on not only conventional distance pruning, keyword pruning, and attribute pruning but also query direction pruning. Particularly, by comparing the position relationship between the query direction and the sector (sector ring) region segmented by the DAPC-based method, the search space that does not meet the query direction is pruned. In addition, we discuss the applicability of our scheme for handling why-not questions on regional spatial keyword queries (SKQ), ordinary direction-aware top- $k$ SKQ queries and complex scoring SKQ queries. Finally, a series of experiments are conducted on two real datasets to show the efficiency of our DAPC-based method.
摘要:
In the literature, most previous studies on English implicit inter-sentence relation recognition only focused on semantic interactions, which could not exploit the syntactic interactive information in Chinese due to its complicated syntactic structure characteristics. In this paper, we propose a novel and effective model DSGCN-RoBERTa to learn the interaction features implied in sentences from both syntactic and semantic perspectives. To generate a rich contextual sentence embedding, we exploit RoBERTa, a large-scale pre-trained language model based on the transformer unit. DSGCN-RoBERTa consists of two key modules, the syntactic interaction and the semantic interaction modules. Specifically, the syntactic interaction module helps capture the depth-level structure information, including non-consecutive words and their relations, while the semantic interaction module enables the model to understand the context from the whole sentence to the local words. Furthermore, on top of such multi-perspective feature representations, we design a strength-dependent matching strategy that is able to adaptively capture the strong relevant interactive information in a fine-grained level. Extensive experiments demonstrate that the proposed method achieved state-of-the-art results on benchmarks Chinese compound sentence corpus CCCS and Chinese discourse corpus CDTB datasets. We also achieve comparable performance on the English corpus PDTB that demonstrates the superiority of our method.
期刊:
Information Sciences,2022年593:505-526 ISSN:0020-0255
通讯作者:
Yanhong Li
作者机构:
[Wang, Pengfei] School of Computer, Central China Normal University, Wuhan, China;Hubei Provincial Key Laboratory of Artificial Intelligence and Smart Learning, CCNU, Wuhan, China;National Language Resources Monitoring and Research Center for Network Media, CCNU, Wuhan, China;[Li, Yanhong] Department of Computer Science, South-Central University for Nationalities, Wuhan, China;[Zheng, Bolong; Li, Guohui] School of Computer of Science and Technology, Huazhong University of Science and Technology, Wuhan, China
通讯机构:
[Yanhong Li] D;Department of Computer Science, South-Central University for Nationalities, Wuhan, China
关键词:
Keyword attribute;SGI Index;Spatial keyword query;Temporal information
作者机构:
[Li, Yuan; Sun, Kaili; Li, Yang] Cent China Normal Univ, Sch Comp Sci, Wuhan 430079, Hubei, Peoples R China.;[Deng, Dunhua] Cent China Normal Univ, Res Ctr Language & Language Educ, Wuhan 430079, Hubei, Peoples R China.
通讯机构:
[Li, Yang] C;Cent China Normal Univ, Sch Comp Sci, Wuhan 430079, Hubei, Peoples R China.
摘要:
Relation classification is a vital task in natural language processing, and it is screening for semantic relation between clauses in texts. This paper describes a study of relation classification on Chinese compound sentences without connectives. There exists an implicit relation in a compound sentence without connectives, which makes it difficult to realize the recognition of relation. The major challenges that relation classification modeling faces are how to obtain the contextual representation of sentence and relation dependence features between clauses. To solve this problem, we propose a novel Inatt-MCNN model to extract sentence features and classify relations by combining multi-channel CNN and Inner-attention mechanism. This network structure utilizes CNN to extract local features of sentences and Inner-attention to capture sentence-level feature representations for this relation classification task. Besides, since the Inner-attention is based on Bi-LSTM, the global and long-term dependence semantic information can be well obtained in Inatt-MCNN to promote the model performance. We conduct experiments on two public Chinese discourse datasets: the Chinese compound sentence corpus (CCCS) dataset and the Tsinghua Chinese Treebank(TCT) dataset. Compared with the previous public methods, Inatt-MCNN model has superior performance and achieves the highest accuracy, especially on the CCCS dataset.
摘要:
A low complexity carrier phase estimation (CPE) algorithm for M-ary quadrature amplitude modulation (m-QAM) optical communication systems is investigated in this paper. In the proposed CPE algorithm, a two-stage CPE method is adopted. In the first stage, the QPSK points of the constellation are picked out to achieve a coarse phase estimation using the traditional Viterbi and Viterbi algorithm. In the second stage, all the points of the constellation are used for a fine phase estimation. In addition, the fourth-power operation is replaced by the 4-level absolute operation for the removal of modulated data phase, which greatly reduced the complexity. The proposed method was investigated through simulation, with 16-QAM, 32-QAM and 64-QAM modulation formats, respectively. The simulation results show that the proposed algorithm has both good linewidth tolerance and amplified spontaneous emission noise tolerance as well as low complexity. Moreover, when the equalization enhanced phase noise is considered, the proposed method also has better performance than traditional algorithm.