原型网络直接应用于小样本命名实体识别(few-shot named entity recognition,FEW-NER)时存在以下问题:非实体之间不具有较强的语义关系,对实体和非实体都采用相同的方式构造原型将会造成非实体原型不能准确表示非实体的语义特征;仅使用...原型网络直接应用于小样本命名实体识别(few-shot named entity recognition,FEW-NER)时存在以下问题:非实体之间不具有较强的语义关系,对实体和非实体都采用相同的方式构造原型将会造成非实体原型不能准确表示非实体的语义特征;仅使用平均实体向量表示作为原型的计算方式将难以捕捉语义特征相差较大的同类实体.针对上述问题,提出基于细粒度原型网络的小样本命名实体识别(FEW-NER based on fine-grained prototypical networks,FNFP)方法,有助于提高小样本命名实体识别的标注效果.首先,为不同的查询集样本构造不同的非实体原型,捕捉句子中关键的非实体语义特征,得到更为细粒度的原型,提升模型对非实体的识别效果;然后,设计一个不一致性度量模块以衡量同类实体之间的不一致性,对实体与非实体采用不同的度量函数,从而减小同类样本之间的特征表示,提升原型的特征表示能力;最后,引入维特比解码器捕捉标签转换关系,优化最终的标注序列.实验结果表明,采用基于细粒度原型网络的小样本命名实体识别方法,在大规模小样本命名实体识别数据集FEW-NERD上,较基线方法获得提升;同时在跨领域数据集上验证所提方法在不同领域场景下的泛化能力.展开更多
Densely deployed Wi Fi networks will play a crucial role in providing the capacity for next generation mobile internet. However, due to increasing interference, overlapped channels in Wi Fi networks and throughput eff...Densely deployed Wi Fi networks will play a crucial role in providing the capacity for next generation mobile internet. However, due to increasing interference, overlapped channels in Wi Fi networks and throughput efficiency degradation, densely deployed Wi Fi networks is not a guarantee to obtain higher throughput. An emergent challenge is how to effi ciently utilize scarce spectrum resources, by matching physical layer resources to traffi c demand. In this aspect, access control allocation strategies play a pivotal role but remain too coarse-grained. As a solution, this research proposes a flexible framework for fine-grained channel width adaptation and multi-channel access in Wi Fi networks. This approach, named SFCA(Subcarrier Fine-grained Channel Access), adopts DOFDM(Discontinuous Orthogonal Frequency Division Multiplexing) at the PHY layer. It allocates the frequency resource with a subcarrier granularity, which facilitates the channel width adaptation for multi-channel access and thus brings more fl exibility and higher frequency efficiency. The MAC layer uses a frequencytime domain backoff scheme, which combines the popular time-domain BEB scheme with a frequency-domain backoff to decrease access collision, resulting in higher access probability for the contending nodes. SFCA is compared with FICA(an established access scheme)showing significant outperformance. Finally we present results for next generation 802.11 ac Wi Fi networks.展开更多
文摘原型网络直接应用于小样本命名实体识别(few-shot named entity recognition,FEW-NER)时存在以下问题:非实体之间不具有较强的语义关系,对实体和非实体都采用相同的方式构造原型将会造成非实体原型不能准确表示非实体的语义特征;仅使用平均实体向量表示作为原型的计算方式将难以捕捉语义特征相差较大的同类实体.针对上述问题,提出基于细粒度原型网络的小样本命名实体识别(FEW-NER based on fine-grained prototypical networks,FNFP)方法,有助于提高小样本命名实体识别的标注效果.首先,为不同的查询集样本构造不同的非实体原型,捕捉句子中关键的非实体语义特征,得到更为细粒度的原型,提升模型对非实体的识别效果;然后,设计一个不一致性度量模块以衡量同类实体之间的不一致性,对实体与非实体采用不同的度量函数,从而减小同类样本之间的特征表示,提升原型的特征表示能力;最后,引入维特比解码器捕捉标签转换关系,优化最终的标注序列.实验结果表明,采用基于细粒度原型网络的小样本命名实体识别方法,在大规模小样本命名实体识别数据集FEW-NERD上,较基线方法获得提升;同时在跨领域数据集上验证所提方法在不同领域场景下的泛化能力.
基金supported by National Natural Science Foundation of China(No.61471376)the 863 project(No.2014AA01A701)
文摘Densely deployed Wi Fi networks will play a crucial role in providing the capacity for next generation mobile internet. However, due to increasing interference, overlapped channels in Wi Fi networks and throughput efficiency degradation, densely deployed Wi Fi networks is not a guarantee to obtain higher throughput. An emergent challenge is how to effi ciently utilize scarce spectrum resources, by matching physical layer resources to traffi c demand. In this aspect, access control allocation strategies play a pivotal role but remain too coarse-grained. As a solution, this research proposes a flexible framework for fine-grained channel width adaptation and multi-channel access in Wi Fi networks. This approach, named SFCA(Subcarrier Fine-grained Channel Access), adopts DOFDM(Discontinuous Orthogonal Frequency Division Multiplexing) at the PHY layer. It allocates the frequency resource with a subcarrier granularity, which facilitates the channel width adaptation for multi-channel access and thus brings more fl exibility and higher frequency efficiency. The MAC layer uses a frequencytime domain backoff scheme, which combines the popular time-domain BEB scheme with a frequency-domain backoff to decrease access collision, resulting in higher access probability for the contending nodes. SFCA is compared with FICA(an established access scheme)showing significant outperformance. Finally we present results for next generation 802.11 ac Wi Fi networks.
