The original online version of this article (Journal of Electronics (China), Vol. 28, No.3, May 2011, pp.389-395; DOI: 10.1007/s11767-011-0549-1) unfortunately contains a mistake on the author affiliation of Page...The original online version of this article (Journal of Electronics (China), Vol. 28, No.3, May 2011, pp.389-395; DOI: 10.1007/s11767-011-0549-1) unfortunately contains a mistake on the author affiliation of Page 389. The correct form is given below:展开更多
Short gate-length High Electron Mobility Transistors (HEMTs) have been observed to exhibit kinks in their drain current-voltage (I-V) characteristics. To model this nonlinear effect, we present an effective approach t...Short gate-length High Electron Mobility Transistors (HEMTs) have been observed to exhibit kinks in their drain current-voltage (I-V) characteristics. To model this nonlinear effect, we present an effective approach that is easily incorporated into most existing empirical HEMT I-V models. This has been done by modifying the channel length modulation parameter to account for the kink effect. Moreover, the definitions of the left parameters in the original model will not be influenced, and the improved HEMT I-V model enhances its bias range of operation for which accuracy is maintained. The proposed modeling method is validated through DC/ Pulsed I-V as well as large-signal power measurements.展开更多
当前,深度主动学习(Deep Active Learning,DAL)在分类数据标注工作中获得成功,但如何筛选出最能提升模型性能的样本仍是难题.本文提出基于弱标签争议的半自动分类数据标注方法(Dispute about Weak Label based Deep Active Learning,DWL...当前,深度主动学习(Deep Active Learning,DAL)在分类数据标注工作中获得成功,但如何筛选出最能提升模型性能的样本仍是难题.本文提出基于弱标签争议的半自动分类数据标注方法(Dispute about Weak Label based Deep Active Learning,DWLDAL),迭代地筛选出模型难以区分的样本,交给人工进行准确标注.该方法包含伪标签生成器和弱标签生成器,伪标签生成器是在准确标注的数据集上训练而成,用于生成无标签数据的伪标签;弱标签生成器则是在带伪标签的随机子集上训练而成.弱标签生成器委员会决定哪些无标签数据最有争议,则交给人工标注.本文针对文本分类问题,在公开数据集IMDB(Internet Movie DataBase)、20NEWS(20NEW Sgroup)和chnsenticorp(chnsenticorp_htl_all)上进行实验验证.从数据标注和分类任务的准确性2个角度,对3种不同投票决策方式进行评估.DWLDAL方法中数据标注的F1分数比现有方法Snuba分别提高30.22%、14.07%和2.57%,DWLDAL方法中分类任务的F1分数比Snuba分别提高1.01%、22.72%和4.83%.展开更多
文摘The original online version of this article (Journal of Electronics (China), Vol. 28, No.3, May 2011, pp.389-395; DOI: 10.1007/s11767-011-0549-1) unfortunately contains a mistake on the author affiliation of Page 389. The correct form is given below:
文摘Short gate-length High Electron Mobility Transistors (HEMTs) have been observed to exhibit kinks in their drain current-voltage (I-V) characteristics. To model this nonlinear effect, we present an effective approach that is easily incorporated into most existing empirical HEMT I-V models. This has been done by modifying the channel length modulation parameter to account for the kink effect. Moreover, the definitions of the left parameters in the original model will not be influenced, and the improved HEMT I-V model enhances its bias range of operation for which accuracy is maintained. The proposed modeling method is validated through DC/ Pulsed I-V as well as large-signal power measurements.
文摘当前,深度主动学习(Deep Active Learning,DAL)在分类数据标注工作中获得成功,但如何筛选出最能提升模型性能的样本仍是难题.本文提出基于弱标签争议的半自动分类数据标注方法(Dispute about Weak Label based Deep Active Learning,DWLDAL),迭代地筛选出模型难以区分的样本,交给人工进行准确标注.该方法包含伪标签生成器和弱标签生成器,伪标签生成器是在准确标注的数据集上训练而成,用于生成无标签数据的伪标签;弱标签生成器则是在带伪标签的随机子集上训练而成.弱标签生成器委员会决定哪些无标签数据最有争议,则交给人工标注.本文针对文本分类问题,在公开数据集IMDB(Internet Movie DataBase)、20NEWS(20NEW Sgroup)和chnsenticorp(chnsenticorp_htl_all)上进行实验验证.从数据标注和分类任务的准确性2个角度,对3种不同投票决策方式进行评估.DWLDAL方法中数据标注的F1分数比现有方法Snuba分别提高30.22%、14.07%和2.57%,DWLDAL方法中分类任务的F1分数比Snuba分别提高1.01%、22.72%和4.83%.