模式识别与人工智能
Home      About Journal      Editorial Board      Instructions      Ethics Statement      Contact Us                   中文
Pattern Recognition and Artificial Intelligence
22 Judgement and Disposal of Academic Misconduct Article
22 Copyright Transfer Agreement
22 Proof of Confidentiality
22 Requirements for Electronic Version
More....
22 Chinese Association of Automation
22 National ResearchCenter for Intelligent Computing System
22 Institute of Intelligent Machines,Chinese Academy of Sciences
More....
 
 
2024 Vol.37 Issue.1, Published 2024-01-25
Papers and Reports    Researches and Applications   
   
0
ZHENG Nanning
2024 Vol. 37 (1): 0-0 [Abstract] ( 106 ) [HTML 1KB] [ PDF 181KB] ( 291 )
Papers and Reports
1 Medical Image Segmentation Method with Triplet-Path Network
JIANG Qingting, YE Hailiang, CAO Feilong
Convolutional neural networks make certain progress in medical image segmentation tasks due to their powerful feature extraction capabilities. However, the accuracy of edge segmentation still needs to be improved. To address this problem, a triplet-path network based on edge selection graph reasoning is proposed in this paper, including the target localization path, edge selection path and refinement path. In the target localization path, a multi-scale feature fusion module is designed to aggregate high-level features for the localization of lesion regions. In the edge selection path, an edge-selective graph reasoning module is constructed for edge screening of low-level features and graph reasoning to ensure the edge shape of the relevant lesion region. In the refinement path, a progressive group level refinement module is established to refine the structure information and details of different scale features. Moreover, a composite loss fusing weighted Focal Tversky loss and a weighted intersection over union loss is introduced to mitigate the effects of class imbalance. Experimental results on public datasets demonstrate the superior performance of the proposed method.
2024 Vol. 37 (1): 1-12 [Abstract] ( 277 ) [HTML 1KB] [ PDF 1462KB] ( 462 )
13 Computerized Adaptive Testing Method Based on Reinforcement Learning for Series Diagnosis
LIU Zirui, WU Jinze, YAO Fangzhou, LIU Qi, CHEN Enhong, SHA Jing, WANG Shijin, SU Yu
Computerized adaptive testing is designed to select appropriate questions for students based on their historical performance, thereby measuring their actual ability quickly and effectively. However, in intelligent education scenarios, the existing selection strategy of traditional computerized adaptive testing is still faced with some problems such as target complexity and knowledge sparseness. To solve these problems, a computerized adaptive testing method based on reinforcement learning for series diagnosis is proposed in this paper to accurately assess students' knowledge proficiency for intelligent scenarios. A student simulator and a student portrait model based on series diagnosis model are adopted. To address the complexity of computerized adaptive testing goals in real-world scenarios, five evaluation indicators are designed, including accuracy of weak points, coupling of prediction performance, adaptive testing duration, testing anomaly rate and testing difficulty structure. Furthermore, a selection strategy for reinforcement learning based computerized adaptive testing is proposed. The dual-channel self-attention learning module and the contradiction learning module are utilized to ameliorate knowledge sparseness problem. Experiments on real datasets show that the proposed selection strategy not only efficiently measures students' actual abilities, but also optimizes their answering experience. The selected questions exhibit a certain level of interpretability, and the method provides a feasible solution for computerized adaptive testing in intelligent education scenarios.
2024 Vol. 37 (1): 13-26 [Abstract] ( 134 ) [HTML 1KB] [ PDF 1169KB] ( 334 )
27 Efficient Encrypted Range Query Integrity Authentication for Hundreds of Millions of Records
WANG Zhaokang, PAN Jiahui, ZHOU Lu
The encrypted query integrity authentication mechanism can provide assurance for the reliability of the query results while protecting the data privacy of artificial intelligence applications. However, the existing encrypted range query integrity authentication methods suffer from high overhead in authentication data structure construction and poor data scalability. To address these issues, the causes of performance bottlenecks in secure verifiable and efficient framework(ServeDB)are analyzed. Based on the analysis conclusions, a cube-cell-based authentication method(CubeTree) is proposed for the encrypted range query integrity authentication problem. A quantile-normalization-based data redistribution optimization is adopted to balance the data distribution in the domain. The encoding overheads of data records are reduced by the data redistribution optimization. Furthermore, a flat balanced K-ary tree structure and a cube-cell-based index authentication data structure are proposed. The redundancy of the authentication data structure is significantly reduced by merging data records with same codes and adopting cube cells as the basic units. Consequently, the computational and storage costs of the CubeTree construction are decreased. Experiments on real-world and synthetic datasets show that CubeTree can significantly reduce the construction costs of the authentication data structure and the generation/verification costs of query integrity proofs, while efficiently handling large-scale datasets with hundreds of millions of data records.
2024 Vol. 37 (1): 27-46 [Abstract] ( 103 ) [HTML 1KB] [ PDF 1556KB] ( 256 )
Researches and Applications
47 Research and Prospect of Robotic Surgical System Based on Human-Machine-Environment Information Flow
CUI Haoxin, WANG Rong, ZHENG Nan, ZHANG Song, REN Tong, LIANG Yujing

Robot-assisted surgical systems continue to gain widespread application in various surgical fields, due to their minimally invasive, precise, flexible and tremor-free attributes. However, the advantages of both humans and machines are not fully exploited in the existing robot-assisted surgical systems, and the performance should be improved in the intelligent interaction aspect. Therefore, the development of interactive relationships within robot-assisted surgical systems is analyzed from a systems science perspective and the deficiencies in human-machine interaction are discussed from multiple viewpoints. Then, an "human-machine-environment" information flow framework for robot-assisted surgical systems is constructed. Taking the example of robotic-assisted internal mammary artery acquisition scenarios, the intricate interactions among the "human-machine-environment" different components are illustrated. Finally, based on the theory of "human-machine integration" and the established "human-machine-environment" information flow framework, a design approach for a new generation of robot-assisted surgical systems with the goal of "human-machine integration and intelligent co-development" is proposed. This proposal serves as a valuable reference for realizing the goal of safer and more efficient robotic minimally invasive surgery.
2024 Vol. 37 (1): 47-57 [Abstract] ( 97 ) [HTML 1KB] [ PDF 1516KB] ( 359 )
58 Nonparametric Additive Quantile Regression Model Based on Fused Lasso
FU Manxia, ZHOU Shuisheng
Additive quantile regression provides a flexible and robust method for modeling non-linear relationships. Methods for fitting the additive quantile models rely on spline functions to approximate components. However, the required prior selection of nodes results in slow computation speed and it renders the methods unsuitable for large-scale data problems. Therefore, a nonparametric additive quantile regression model based on the fused Lasso(AQFL) is proposed in this paper. AQFL leverages a compromise between the fused Lasso penalty and the l2 penalty for estimating and selecting variables in the additive quantile regression model. The fused Lasso penalty is employed to make the model compute fast and localize adaptively, thereby achieving the prediction for the desired quantile or even extreme quantiles. Additionally, in combination with the l2 penalty, AQFL compresses the covariate function values with a small impact on the response to zero in high-dimensional data, thereby achieving variable selection. Furthermore, a block coordinate alternating direction method of multipliers(BC-ADMM) algorithm is presented to ensure convergence to the global optimum and demonstrate the prediction consistency of AQFL. Experimental results on synthetic data and ground pork data demonstrate the superiority of AQFL in prediction accuracy and robustness.
2024 Vol. 37 (1): 58-72 [Abstract] ( 91 ) [HTML 1KB] [ PDF 764KB] ( 262 )
73 Lipreading Based on Multiple Visual Attention
XIE Yincen, XUE Feng, CAO Mingwei
Lipreading is a technology that translates the silent video of a single speaker's lip motion into text. Due to the small amplitude of lip movements, the feature differentiation ability and the generalization ability of the model are both weak. To address this issue, the purification of lipreading visual features is studied from three dimensions including time, space and channel. A method for lipreading based on multiple visual attention network(LipMVA) is proposed. Firstly, channel-level features are calibrated adaptively by channel attention to mitigate the interference from meaningless channels. Then, two spatio-temporal attention modules with different granularities are employed to suppress the effect of unimportant pixels or frames. Finally, experiments on CMLR and GRID datasets demonstrate LipMVA can reduce the error rate and therefore its effectiveness is verified.
2024 Vol. 37 (1): 73-84 [Abstract] ( 180 ) [HTML 1KB] [ PDF 2836KB] ( 421 )
85 Multi-input Fusion Spelling Error Correction Model Based on Contrast Optimization
WU Yaoyao, HUANG Ruizhang, BAI Ruina, CAO Junhang, ZHAO Jianhui
Chinese spelling correction is essential in text editing. Most of the existing Chinese spelling error correction models are single input models, and there are limitations in the semantic information and error correction results of the models. In this paper, a multi-input fusion spelling error correction method based on contrast optimization, MIF-SECCO, is proposed. MIF-SECCO contains two stages: multi-input semantic learning and contrast learning-driven semantic fusion error correction. In the first stage, preliminary error correction results from multiple single input models are integrated to provide sufficient complementary semantic information for semantic fusion. In the second stage, multiple complementary sentence semantics are optimized based on the contrastive learning approach to avoid over-correction of sentences by the model. The limitations of error correction results of the model are improved by fusing multiple complementary semantics for re-correction of erroneous sentences. Experimental results on the public datasets SIGHAN13, SIGHAN14 and SIGHAN15 demonstrate MIF-SECCO effectively improves the error correction performance of the model.
2024 Vol. 37 (1): 85-94 [Abstract] ( 119 ) [HTML 1KB] [ PDF 835KB] ( 294 )
模式识别与人工智能
 

Supervised by
China Association for Science and Technology
Sponsored by
Chinese Association of Automation
NationalResearchCenter for Intelligent Computing System
Institute of Intelligent Machines, Chinese Academy of Sciences
Published by
Science Press
 
Copyright © 2010 Editorial Office of Pattern Recognition and Artificial Intelligence
Address: No.350 Shushanhu Road, Hefei, Anhui Province, P.R. China Tel: 0551-65591176 Fax:0551-65591176 Email: bjb@iim.ac.cn
Supported by Beijing Magtech 　Email:support@magtech.com.cn