Productivity potentials of upland rice landraces (URLs) are continuously compromised by scanty water supply due to competing priorities for irrigations and fluctuations in agro-ecological conditions peculiar to low-in...Productivity potentials of upland rice landraces (URLs) are continuously compromised by scanty water supply due to competing priorities for irrigations and fluctuations in agro-ecological conditions peculiar to low-input farming systems. A growing demand for rice amidst decline in productive agricultural areas plunges into an urgent contribution of marginal areas critical in attaining food sufficiency among Filipino households. Agronomic performances of URLs locally found in Catanduanes province, Philippines were evaluated in a replicated trial utilizing three URLs: Kamanang [1], Kadari [2] and Bulaw [3]. The experimental set-up was closely monitored for consistent dry moisture content and zero fertilizer application. Results were significant for traditional upland landraces: Kamanang and Kadari, scored in terms of higher germination rate, increment plant height and the number of tillers during the vegetative stage. Leaf color gradations, although statistically not significant across cultivars, were suggestive of varying adaptive performance between traditional cultivars subjected to low-input system. Putative low-input responsive lines indicated by the higher survival percentage and overall agronomic responses were selected from the study site. Screened lines took part of the advance population which would be potentially able to tolerate poor soil conditions (i.e. poor storehouse of water and nutrients) especially in areas with edaphological constraints and changing rainfall distribution pattern. The development of putative drought tolerant lines among URLs forms the most economical solution implicated to areas with limited access to agricultural interventions.展开更多
现有多视角聚类算法存在:1)在学习低维表征的过程中无法准确捕获或忽略嵌入在多视角数据中的高阶信息和互补信息;2)未能准确捕获数据局部信息;3)信息捕获方法缺少对噪声点鲁棒性等问题.为解决上述问题,提出一种自适应张量奇异值收缩的...现有多视角聚类算法存在:1)在学习低维表征的过程中无法准确捕获或忽略嵌入在多视角数据中的高阶信息和互补信息;2)未能准确捕获数据局部信息;3)信息捕获方法缺少对噪声点鲁棒性等问题.为解决上述问题,提出一种自适应张量奇异值收缩的多视角聚类(multi-view clustering based on adaptive tensor singular value shrinkage,ATSVS)算法.ATSVS首先提出一种符合秩特性的张量对数行列式函数对表示张量施加低秩约束,在张量奇异值分解(tensor singular value decomposition,t-SVD)过程中能够根据奇异值自身大小进行自适应收缩,更加准确地进行张量秩估计,进而从全局角度精准捕获多视角数据的高阶信息和互补信息.然后采用一种结合稀疏表示和流形正则技术优势的l_(1,2)范数捕获数据的局部信息,并结合l_(2,1)范数对噪声施加稀疏约束,提升算法对噪声点的鲁棒性.与11个对比算法在9个数据集上的实验结果显示,ATSVS的聚类性能均优于其他对比算法.因此,ATSVS是一个能够有效处理多视角数据聚类任务的优秀算法.展开更多
区域内的低品位热源,如低温工业余热、数据中心的废热以及生活污水中的余热等未被充分利用的热能资源,具有显著的区域供热潜力,配置分布式电热泵单元可提升热源温度品位、实现低成本供热。为促进区域协同供能并降低光伏随机波动影响,该...区域内的低品位热源,如低温工业余热、数据中心的废热以及生活污水中的余热等未被充分利用的热能资源,具有显著的区域供热潜力,配置分布式电热泵单元可提升热源温度品位、实现低成本供热。为促进区域协同供能并降低光伏随机波动影响,该文考虑热泵灵活性特征,针对区域低品位热源、光伏与热电联产单元多种供能形式的协同运行与收益分配策略展开研究。首先,考虑多时段光伏预测随机误差与电热泵拟合模型误差,建立电热泵与热电联产机组基于仿射策略的日前机会约束优化模型与日内优化模型;其次,提出基于日前、日内贡献度权重系数的非对称纳什议价方法来制定各主体收益分配策略;最后,利用改进自适应交替方向乘子法(alternating direction multiplier method,ADMM)来解决区域内各主体之间的收益分配问题。算例结果表明,所提策略在实现含低品位热源区域协同优化运行的同时,有效降低了光伏与热泵拟合模型误差的影响,并提供了区域内考虑多时间尺度贡献度的合理收益分配方案。展开更多
情感识别是人机交互智能化的关键环节.脑电(ElectroEncephaloGram,EEG)信号因其蕴含丰富的生物信息且难以伪装,成为情感分析的重要载体.然而,EEG信号特征复杂多变,且存在显著的个体间差异和时变性,导致传统机器学习方法的情感分类准确...情感识别是人机交互智能化的关键环节.脑电(ElectroEncephaloGram,EEG)信号因其蕴含丰富的生物信息且难以伪装,成为情感分析的重要载体.然而,EEG信号特征复杂多变,且存在显著的个体间差异和时变性,导致传统机器学习方法的情感分类准确率低、泛化能力差.针对这一挑战,本文提出了一种基于重构迁移子空间多视角领域适应(Reconstructed Transfer Subspace based Multi-View Domain Adaptation,RTS-MVDA)方法.该方法将不同特征视为独立视角,通过多视角学习探索各视角的独特性和重要性,并探索其互补关系.其核心在于将源域与目标域的多视角数据投影到一个带有低秩约束的重构迁移子空间.在该子空间中,RTS-MVDA一方面利用重构项恢复原始数据信息,并通过低秩表示保留主要判别信息;另一方面,RTS-MVDA实施线性变换对齐源域和目标域,减少领域间的分布差异.此外,RTS-MVDA构建多视角监督判别项和全局结构保持项,多视角监督判别项利用源域标签信息增强类内紧凑性和类间分离性,全局结构保持项保持数据在迁移子空间中的全局结构分布,从而更有效地将源域的判别知识迁移至目标域.在公开DEAP(Database for Emotion Analysis using Physiological signals)数据集上的实验验证表明:所提RTS-MVDA方法在唤醒度和效价维度上分别达到了73.15%和72.91%的平均准确率,其Precision、Recall和F1-score指标均显著优于相关对比方法,有效提升了跨被试EEG情感识别的准确性和泛化能力.展开更多
文摘Productivity potentials of upland rice landraces (URLs) are continuously compromised by scanty water supply due to competing priorities for irrigations and fluctuations in agro-ecological conditions peculiar to low-input farming systems. A growing demand for rice amidst decline in productive agricultural areas plunges into an urgent contribution of marginal areas critical in attaining food sufficiency among Filipino households. Agronomic performances of URLs locally found in Catanduanes province, Philippines were evaluated in a replicated trial utilizing three URLs: Kamanang [1], Kadari [2] and Bulaw [3]. The experimental set-up was closely monitored for consistent dry moisture content and zero fertilizer application. Results were significant for traditional upland landraces: Kamanang and Kadari, scored in terms of higher germination rate, increment plant height and the number of tillers during the vegetative stage. Leaf color gradations, although statistically not significant across cultivars, were suggestive of varying adaptive performance between traditional cultivars subjected to low-input system. Putative low-input responsive lines indicated by the higher survival percentage and overall agronomic responses were selected from the study site. Screened lines took part of the advance population which would be potentially able to tolerate poor soil conditions (i.e. poor storehouse of water and nutrients) especially in areas with edaphological constraints and changing rainfall distribution pattern. The development of putative drought tolerant lines among URLs forms the most economical solution implicated to areas with limited access to agricultural interventions.
文摘现有多视角聚类算法存在:1)在学习低维表征的过程中无法准确捕获或忽略嵌入在多视角数据中的高阶信息和互补信息;2)未能准确捕获数据局部信息;3)信息捕获方法缺少对噪声点鲁棒性等问题.为解决上述问题,提出一种自适应张量奇异值收缩的多视角聚类(multi-view clustering based on adaptive tensor singular value shrinkage,ATSVS)算法.ATSVS首先提出一种符合秩特性的张量对数行列式函数对表示张量施加低秩约束,在张量奇异值分解(tensor singular value decomposition,t-SVD)过程中能够根据奇异值自身大小进行自适应收缩,更加准确地进行张量秩估计,进而从全局角度精准捕获多视角数据的高阶信息和互补信息.然后采用一种结合稀疏表示和流形正则技术优势的l_(1,2)范数捕获数据的局部信息,并结合l_(2,1)范数对噪声施加稀疏约束,提升算法对噪声点的鲁棒性.与11个对比算法在9个数据集上的实验结果显示,ATSVS的聚类性能均优于其他对比算法.因此,ATSVS是一个能够有效处理多视角数据聚类任务的优秀算法.
文摘区域内的低品位热源,如低温工业余热、数据中心的废热以及生活污水中的余热等未被充分利用的热能资源,具有显著的区域供热潜力,配置分布式电热泵单元可提升热源温度品位、实现低成本供热。为促进区域协同供能并降低光伏随机波动影响,该文考虑热泵灵活性特征,针对区域低品位热源、光伏与热电联产单元多种供能形式的协同运行与收益分配策略展开研究。首先,考虑多时段光伏预测随机误差与电热泵拟合模型误差,建立电热泵与热电联产机组基于仿射策略的日前机会约束优化模型与日内优化模型;其次,提出基于日前、日内贡献度权重系数的非对称纳什议价方法来制定各主体收益分配策略;最后,利用改进自适应交替方向乘子法(alternating direction multiplier method,ADMM)来解决区域内各主体之间的收益分配问题。算例结果表明,所提策略在实现含低品位热源区域协同优化运行的同时,有效降低了光伏与热泵拟合模型误差的影响,并提供了区域内考虑多时间尺度贡献度的合理收益分配方案。
文摘情感识别是人机交互智能化的关键环节.脑电(ElectroEncephaloGram,EEG)信号因其蕴含丰富的生物信息且难以伪装,成为情感分析的重要载体.然而,EEG信号特征复杂多变,且存在显著的个体间差异和时变性,导致传统机器学习方法的情感分类准确率低、泛化能力差.针对这一挑战,本文提出了一种基于重构迁移子空间多视角领域适应(Reconstructed Transfer Subspace based Multi-View Domain Adaptation,RTS-MVDA)方法.该方法将不同特征视为独立视角,通过多视角学习探索各视角的独特性和重要性,并探索其互补关系.其核心在于将源域与目标域的多视角数据投影到一个带有低秩约束的重构迁移子空间.在该子空间中,RTS-MVDA一方面利用重构项恢复原始数据信息,并通过低秩表示保留主要判别信息;另一方面,RTS-MVDA实施线性变换对齐源域和目标域,减少领域间的分布差异.此外,RTS-MVDA构建多视角监督判别项和全局结构保持项,多视角监督判别项利用源域标签信息增强类内紧凑性和类间分离性,全局结构保持项保持数据在迁移子空间中的全局结构分布,从而更有效地将源域的判别知识迁移至目标域.在公开DEAP(Database for Emotion Analysis using Physiological signals)数据集上的实验验证表明:所提RTS-MVDA方法在唤醒度和效价维度上分别达到了73.15%和72.91%的平均准确率,其Precision、Recall和F1-score指标均显著优于相关对比方法,有效提升了跨被试EEG情感识别的准确性和泛化能力.
