针对人类示教轨迹样本存在的时间和空间不对齐导致难以提取运动特征的问题,首先提出了基于典型时间规整(Canonical Time Warping,CTW)算法用于多条轨迹对齐的方法,并将其引入到软-动态时间规整(soft-dynamic time warping,soft-DTW)算...针对人类示教轨迹样本存在的时间和空间不对齐导致难以提取运动特征的问题,首先提出了基于典型时间规整(Canonical Time Warping,CTW)算法用于多条轨迹对齐的方法,并将其引入到软-动态时间规整(soft-dynamic time warping,soft-DTW)算法中以提取轨迹模板,其次在CTW算法中引入了一个新的变量,以提升CTW算法在对齐多条轨迹方面的能力;最后,在实验中利用多种轨迹验证了所提出的轨迹模板提取方法,实验结果表明所提出的方法可以从人类示教轨迹中快速地提取共有的运动特征,并且对示教轨迹在时间和空间上的差异具有较好的鲁棒性.展开更多
语音驱动3D面部运动的研究主要聚焦于拓展多说话人的3D面部运动数据以及获取高质量音频特征上,但采集3D面部运动数据往往需要高昂的成本和繁琐的标注流程,单一说话人的少量数据样本又会导致模型因为数据的稀疏性难以获取高质量音频特征...语音驱动3D面部运动的研究主要聚焦于拓展多说话人的3D面部运动数据以及获取高质量音频特征上,但采集3D面部运动数据往往需要高昂的成本和繁琐的标注流程,单一说话人的少量数据样本又会导致模型因为数据的稀疏性难以获取高质量音频特征。针对该问题,论文从时间序列任务中获得启发,将可微动态时间规整(Smoothed formulation of Dynamic Time Warping, Soft-DTW)应用到语音特征与面部网格(Mesh)顶点的跨模态对齐中。经过实验表明,采用Soft-DTW作为损失函数在生成面部动画的唇形吻合度方面相较于使用均方误差(Mean Squared Error, MSE)时有所提高,可以合成更高质量的面部动画。展开更多
The ongoing energy transition,essential for mitigating global warming,stands to benefit significantly from advances in building energy consumption prediction.With the rise of big data,data-driven models have become in...The ongoing energy transition,essential for mitigating global warming,stands to benefit significantly from advances in building energy consumption prediction.With the rise of big data,data-driven models have become increasingly effective in forecasting,with machine learning emerging as the most efficient method for constructing these predictive models.While previous reviews have typically listed various machine learning models for energy consumption prediction,they have often lacked a theoretical perspective explaining why certain models are suitable for different aspects of this domain.In contrast,this review introduces machine learning techniques based on their application phases,covering preprocessing techniques such as feature selection,extraction,and clustering,as well as state-of-the-art predictive models.We provide a comparative theoretical analysis of various models,examining their strengths,weaknesses,and suitability for different forecasting tasks.Additionally,we discuss spatial-temporal considerations in energy consumption forecasting,including the role of Graph Neural Networks and multitask learning.Furthermore,we address a significant challenge in the field,the difficulty of accurately predicting high-fluctuation electricity consumption,and propose potential solutions to tackle this issue.展开更多
文摘针对人类示教轨迹样本存在的时间和空间不对齐导致难以提取运动特征的问题,首先提出了基于典型时间规整(Canonical Time Warping,CTW)算法用于多条轨迹对齐的方法,并将其引入到软-动态时间规整(soft-dynamic time warping,soft-DTW)算法中以提取轨迹模板,其次在CTW算法中引入了一个新的变量,以提升CTW算法在对齐多条轨迹方面的能力;最后,在实验中利用多种轨迹验证了所提出的轨迹模板提取方法,实验结果表明所提出的方法可以从人类示教轨迹中快速地提取共有的运动特征,并且对示教轨迹在时间和空间上的差异具有较好的鲁棒性.
文摘语音驱动3D面部运动的研究主要聚焦于拓展多说话人的3D面部运动数据以及获取高质量音频特征上,但采集3D面部运动数据往往需要高昂的成本和繁琐的标注流程,单一说话人的少量数据样本又会导致模型因为数据的稀疏性难以获取高质量音频特征。针对该问题,论文从时间序列任务中获得启发,将可微动态时间规整(Smoothed formulation of Dynamic Time Warping, Soft-DTW)应用到语音特征与面部网格(Mesh)顶点的跨模态对齐中。经过实验表明,采用Soft-DTW作为损失函数在生成面部动画的唇形吻合度方面相较于使用均方误差(Mean Squared Error, MSE)时有所提高,可以合成更高质量的面部动画。
文摘The ongoing energy transition,essential for mitigating global warming,stands to benefit significantly from advances in building energy consumption prediction.With the rise of big data,data-driven models have become increasingly effective in forecasting,with machine learning emerging as the most efficient method for constructing these predictive models.While previous reviews have typically listed various machine learning models for energy consumption prediction,they have often lacked a theoretical perspective explaining why certain models are suitable for different aspects of this domain.In contrast,this review introduces machine learning techniques based on their application phases,covering preprocessing techniques such as feature selection,extraction,and clustering,as well as state-of-the-art predictive models.We provide a comparative theoretical analysis of various models,examining their strengths,weaknesses,and suitability for different forecasting tasks.Additionally,we discuss spatial-temporal considerations in energy consumption forecasting,including the role of Graph Neural Networks and multitask learning.Furthermore,we address a significant challenge in the field,the difficulty of accurately predicting high-fluctuation electricity consumption,and propose potential solutions to tackle this issue.
