基于预训练和注意机制的意图分类和语义槽填充,提出一种结合双向长短时记忆(bidirectional long short-term memory,BiLSTM)、条件随机场(conditional random fields,CRF)和注意机制的双向编码(bidirectional encoder representations f...
详细信息
基于预训练和注意机制的意图分类和语义槽填充,提出一种结合双向长短时记忆(bidirectional long short-term memory,BiLSTM)、条件随机场(conditional random fields,CRF)和注意机制的双向编码(bidirectional encoder representations from transformers,BERT)具有双向编码表示和注意机制的联合模型。该模型无需过多依赖手工标签数据和领域特定的知识或资源,避免了目前普遍存在的弱泛化能力。在自主公交信息查询系统语料库上进行的试验表明,该模型意图分类的准确性和语义槽填充F1值分别达到98%和96.3%,均产生有效改进。
A direct tunneling model through gate dielectric s in CMOS devices in the frame of WKB approximation is *** the model,an im proved one-band effective mass approximation is used for the hole quantization, where valenc...
详细信息
A direct tunneling model through gate dielectric s in CMOS devices in the frame of WKB approximation is *** the model,an im proved one-band effective mass approximation is used for the hole quantization, where valence band mixing is taken into *** comparing to the experiments, the model is demonstrated to be applicable to both electron and hole tunneling c urrents in CMOS *** effect of the dispersion in oxide energy gap on the tunneling current is also *** model can be further extended to study th e direct tunneling current in future high-k materials.
暂无评论