ID	原文	译文
3773	为了兼顾全局特征和局部特征，同时实现特征的多细粒度表示，构建了多池化融合网络。	To describe both global and local features and implement multiple fine-grained repre-sentations, a multi-pool fusion network was constructed.
3774	为了使监督网络能提取有用的行人前景特征，构建了特征级有监督背景消除网络。	To supervise the network to extract useful foreground features, afeature-level supervised background elimination network was constructed.
3775	采用结合行人分类损失及特征激活损失的多任务学习损失函数，在 3 个公开行人再识别数据集上对方法进行评估，当 MSMT17 作为训练集时，Market-1501 上的跨数据集识别性能 mAP 为 35.53%，相比 ResNet50 网络提升了 9.24%；	The final network loss function was defined asa multi-task loss, which combined both person classification loss and feature activation loss. Three person re-ID bench-marks were employed to evaluate the proposed method. Using MSMT17 as the training set, the cross-dataset mAP forMarket-1501 was 35.53%, which was 9.24% higher than ResNet50.
3776	DukeMTMC-reID 上的跨数据集识别性能 mAP 为 41.45%，相比于 ResNet50 网络提升了 10.72%。	Using MSMT17 as the training set, the cross-datasetmAP for DukeMTMC-reID was 41.45%, which was 10.72% higher than ResNet50.
3777	与现有方法相比，所提方法具有更优的跨数据集行人再识别性能。	Compared with existing methods, the proposed method shows better cross-dataset person re-ID performance.
3778	针对当前网络安全领域知识获取中所依赖的流水线模式存在实体识别错误的传播，未考虑实体识别与关系抽取任务间的联系，以及模型训练缺乏标签语料的问题，提出一种融合对抗主动学习的端到端网络安全知识三元组抽取方法。	Aiming at the problem that using pipeline methods for extracting cybersecurity knowledge triples may causethe errors propagation of entity recognition and did not consider the correlation between entity recognition and relationextraction, and training triple extraction model lacked labeled corpora, an end-to-end cybersecurity knowledge triple ex-traction method with adversarial active learning was proposed.
3779	首先，将实体识别与关系抽取通过联合标注策略建模为序列标注任务；	For knowledge triple extraction, the conventional entityrecognition and relation extraction were modelled as sequence labeling task through joint labeling strategy firstly.
3780	然后，设计融合动态注意力机制的 BiLSTM-LSTM 模型实现实体与关系的联合抽取，并形成三元组；	And then, a BiLSTM-LSTM-based model with dynamic attention mechanism was designed to jointly extract entities and rela-tions, forming triples.
3781	最后，基于对抗网络训练一个判别器模型，增量地筛选出高质量的待标注数据进行标注，并通过迭代训练不断提升联合抽取模型的性能。	Finally, with adversarial learning framework, a discriminator was trained to incrementally selecthigh-quality samples for labeling, and the performance of the joint extraction model was continuously enhanced by itera-tive retraining.
3782	通过实验表明，所提方案中实体−关系联合抽取模型优于现有的网络安全知识抽取方案，并验证了对抗主动学习方法的有效性。	Experiments show that the proposed joint extraction model outperforms the existing cybersecurityknowledge triple extraction methods, and demonstrate the effectiveness of proposed adversarial active learning scheme.