计算机科学 ›› 2022, Vol. 49 ›› Issue (5): 64-70.doi: 10.11896/jsjkx.210400176
成科扬, 王宁, 崔宏纲, 詹永照
CHENG Ke-yang, WANG Ning, CUI Hong-gang, ZHAN Yong-zhao
摘要： 目前,深度学习模型已被广泛部署于各个工业领域。然而,深度学习模型具有的复杂性与不可解释性已成为其应用于高风险领域最主要的瓶颈。在深度学习模型可解释性方法中,最重要的方法是可视化解释方法,其中注意力图是可视化解释方法的主要表现方式,可通过对样本图像中的决策区域进行标注,来直观地展示模型决策依据。目前已有的基于注意力图的可视化解释方法中,单一模型注意力图存在标注区域易出现标注错误而造成可视化可解释性置信度不足的问题。针对上述问题,文中提出了一种基于局部注意力图互迁移的可解释性优化方法,用于提升模型注意力图的标注准确度,展示出精准的决策区域,加强视觉层面对模型决策依据的可解释性。具体表现为:采用轻量模型构建互迁移网络结构,于单一模型层间提取特征图并进行叠加,对全局注意力图进行局部划分,使用皮尔逊相关系数对模型间对应的局部注意力图进行相似度度量,随后将局部注意力图进行正则化并结合交叉熵函数对模型注意力图进行迁移。实验结果表明,所提算法显著提升了模型注意力图标注的准确性,并分别实现了28.2%的平均下降率和29.5%的平均增长率,与最先进的算法相比,其在平均下降率方面实现了3.3%的提升。实验结果表明,所提算法能成功地找出样本图像中预测标签最相关区域,而不局限于视觉可视化区域;与现有的同类方法相比,所提方法能更准确地揭示原始CNN模型的决策依据。
中图分类号:
| [1]YUAN X X,WU Q.Object detection in remote sensing images based on saliency feature and angle information[J].Computer Science,2021,48(4):174-179. [2]YADAV A,VISHWAKARMA D K.Sentiment analysis usingdeep learning architectures:a review[J].Artificial Intelligence Review,2020,53(6):4335-4385. [3]ZEILER M D,FERGUS R.Visualizing and understanding convolutional networks[C]//Proceedings of European Conference on Computer Vision.Berlin:Springer,2014:818-833. [4]GAUR M,FALDUK,SHETH A.Semantics of the black-box:can knowledge graphs help make deep learning systems more interpretable and explainable?[J].IEEE Internet Computing,2021,25(1):51-59. [5]DENIL M,BAZZANI L,LAROCHELLE H,et al.Learningwhere to attend with deep architectures for image tracking[J].Neural Computation,2012,24(8):2151-2184. [6]SELVARAJU R R,COGSWELL M,DAS A,et al.Grad-cam:Visual explanations from deep networks via gradient-based localization[C]//Proceedings of the IEEE International Conference on Computer Vision.2017:618-626. [7]ZHOU B,KHOSLA A,LAPEDRIZA A,et al.Learning deep features for discriminative localization[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2016:2921-2929. [8]ZENG T,ACUNA D E.Modeling citation worthiness by using attention-based bidirectional long short-term memory networks and interpretable models[J].Scientometrics,2020,5(1):1-30. [9]HINTON G,VINYALS O,DEAN J.Distilling the knowledge in a neural network[J].Computer Science,2015,14(7):38-39. [10]YIM J,JOO D,BAE J,et al.A gift from knowledge distillation:fast optimization,network minimization and transfer learning[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2017:4133-4141. [11]HSU E Y,LIU C L,TSENG V S.Multivariate time series early classification with interpretability using deep learning and attention mechanism[M].Advances in Knowledge Discovery and Data Mining.2019. [12]LIU Q,MUKHOPADHYAY S.Unsupervised learning usingpretrained CNN and associative memory bank[C]//2018 International Joint Conference on Neural Networks (IJCNN).IEEE,2018:1-8. [13]PETSIUK V,DAS A,SAENKO K.Rise:randomized input sampling for explanation of black-box models[EB/OL].(2018-09-25)[2018-09-25].https://arxiv.org/abs/1806.07421. [14]SELVARAJU R R,COGSWELL M,DAS A,et al.Grad-CAM:visual explanations from deep networks via gradient-based loca-lization[J].International Journal of Computer Vision,2020,128(2):336-359. [15]CHATTOPADHYAY A,SARKAR A,HOWLADER P,et al.Grad-cam++:Generalized gradient-based visual explanations for deep convolutional networks[C]//2018 IEEE Winter Conference on Applications of Computer Vision (WACV).IEEE,2018:839-847. [16]WANG H,WANG Z,DU M,et al.Score-CAM:Score-weighted visual explanations for convolutional neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops.2020:24-25. [17]ZHANG Y,XIANG T,HOSPEDALES T M,et al.Deep mutual learning[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2018:4320-4328. [18]LIU Q,MUKHOPADHYAY S.Unsupervised learning usingpretrained CNN and associative memory bank[C]//2018 Interational Joint Conference on Neural Networks (IJCNN).IEEE,2018:1-8. [19]CHEN W C,CHANG C C,LEE C R.Knowledge distillationwith feature maps for image classification[C]//Asian Conference on Computer Vision.Cham:Springer,2018:200-215. |
| [1] | 王明, 武文芳, 王大玲, 冯时, 张一飞. 生成链接树:一种高数据真实性的反事实解释生成方法 Generative Link Tree:A Counterfactual Explanation Generation Approach with High Data Fidelity 计算机科学, 2022, 49(9): 33-40. https://doi.org/10.11896/jsjkx.220300158 |
| [2] | 柴慧敏, 张勇, 方敏. 基于特征相似度聚类的空中目标分群方法 Aerial Target Grouping Method Based on Feature Similarity Clustering 计算机科学, 2022, 49(9): 70-75. https://doi.org/10.11896/jsjkx.210800203 |
| [3] | 吴子仪, 李邵梅, 姜梦函, 张建朋. 基于自注意力模型的本体对齐方法 Ontology Alignment Method Based on Self-attention 计算机科学, 2022, 49(9): 215-220. https://doi.org/10.11896/jsjkx.210700190 |
| [4] | 李斌, 万源. 基于相似度矩阵学习和矩阵校正的无监督多视角特征选择 Unsupervised Multi-view Feature Selection Based on Similarity Matrix Learning and Matrix Alignment 计算机科学, 2022, 49(8): 86-96. https://doi.org/10.11896/jsjkx.210700124 |
| [5] | 曾志贤, 曹建军, 翁年凤, 蒋国权, 徐滨. 基于注意力机制的细粒度语义关联视频-文本跨模态实体分辨 Fine-grained Semantic Association Video-Text Cross-modal Entity Resolution Based on Attention Mechanism 计算机科学, 2022, 49(7): 106-112. https://doi.org/10.11896/jsjkx.210500224 |
| [6] | 王毅, 李政浩, 陈星. 基于用户场景的Android 应用服务推荐方法 Recommendation of Android Application Services via User Scenarios 计算机科学, 2022, 49(6A): 267-271. https://doi.org/10.11896/jsjkx.210700123 |
| [7] | 赵璐, 袁立明, 郝琨. 多示例学习算法综述 Review of Multi-instance Learning Algorithms 计算机科学, 2022, 49(6A): 93-99. https://doi.org/10.11896/jsjkx.210500047 |
| [8] | 黄少滨, 孙雪薇, 李熔盛. 基于跨句上下文信息的神经网络关系分类方法 Relation Classification Method Based on Cross-sentence Contextual Information for Neural Network 计算机科学, 2022, 49(6A): 119-124. https://doi.org/10.11896/jsjkx.210600150 |
| [9] | 陈壮, 邹海涛, 郑尚, 于化龙, 高尚. 基于用户覆盖及评分差异的多样性推荐算法 Diversity Recommendation Algorithm Based on User Coverage and Rating Differences 计算机科学, 2022, 49(5): 159-164. https://doi.org/10.11896/jsjkx.210300263 |
| [10] | 朝乐门, 王锐. 数据科学平台:特征、技术及趋势 Data Science Platform:Features,Technologies and Trends 计算机科学, 2021, 48(8): 1-12. https://doi.org/10.11896/jsjkx.210600033 |
| [11] | 王胜, 张仰森, 陈若愚, 向尕. 基于细粒度差异特征的文本匹配方法 Text Matching Method Based on Fine-grained Difference Features 计算机科学, 2021, 48(8): 60-65. https://doi.org/10.11896/jsjkx.200700008 |
| [12] | 王春静, 刘丽, 谭艳艳, 张化祥. 基于模糊颜色特征和模糊相似度的图像检索方法 Image Retrieval Method Based on Fuzzy Color Features and Fuzzy Smiliarity 计算机科学, 2021, 48(8): 191-199. https://doi.org/10.11896/jsjkx.200800202 |
| [13] | 郭奕杉, 刘漫丹. 基于时空轨迹数据的异常检测 Anomaly Detection Based on Spatial-temporal Trajectory Data 计算机科学, 2021, 48(6A): 213-219. https://doi.org/10.11896/jsjkx.201100193 |
| [14] | 孙振强, 罗永龙, 郑孝遥, 章海燕. 一种融合用户情感与相似度的智能旅游路径推荐方法 Intelligent Travel Route Recommendation Method Integrating User Emotion and Similarity 计算机科学, 2021, 48(6A): 226-230. https://doi.org/10.11896/jsjkx.200900119 |
| [15] | 邵超, 宋淑米. 基于信任关系下用户兴趣偏好的协同过滤推荐算法 Collaborative Filtering Recommendation Algorithm Based on User Preference Under Trust Relationship 计算机科学, 2021, 48(6A): 240-245. https://doi.org/10.11896/jsjkx.200700113 |
|
||
首页