目的生成式和传统人工智能模型是信息时代的关键工具。在这些技术的助力下,眼动过程的样本生成与识别显得尤为关键,它已成为深入研究认知机制的重要手段。为了推动生成式人工智能在眼动技术领域的应用发展,解决眼动样本生成及因网络深...目的生成式和传统人工智能模型是信息时代的关键工具。在这些技术的助力下,眼动过程的样本生成与识别显得尤为关键,它已成为深入研究认知机制的重要手段。为了推动生成式人工智能在眼动技术领域的应用发展,解决眼动样本生成及因网络深度增加而导致的不透明性和不可解释性问题,并深入挖掘与幼儿语言发展相关的眼动数据,方法采集4~6岁幼儿理解不同焦点结构的眼动数据,采用生成式人工智能模型-变分自编码器(variational autoencoder,VAE)和传统模型-多层感知器(multi-layer perceptron,MLP)识别眼动模式的发展差异并尝试生成新样本,基于灰色关联分析和混淆矩阵对生成式数据集进行解释。结果结果表明:(1)VAE生成的4岁组、5岁组和6岁组幼儿眼动数据集精度高于MINIST数据集(mixed National Institute of Standards and Technology database),且与MLP分析结果一致,具有准确性、多样性和一定的可解释性;(2)生成式眼动数据及混淆矩阵结果表明,在无焦点结构句式中,幼儿在4~5岁、5~6岁两个阶段理解水平均有提升,而宾语焦点结构和主语焦点结构的眼动特征在4~5岁变化较小,5~6岁变化较大,说明幼儿对焦点结构的理解在5岁是一个关键期,这符合幼儿焦点结构理解发展规律。结论提出的人工智能耦合分析方法,具备有效识别眼动特征发展模式的能力,并能据此生成可靠的新样本。这一方法不仅为生成式人工智能与眼动技术的融合开辟了新的途径,而且为复杂语言理解问题提供了全新的思考方向。展开更多
In order to achieve the functional requirements of multi-moving state, a new autonomous underwater vehicle(AUV) provided with the functions such as the submarine vectorial thrust, landing on the sea bottom, wheel driv...In order to achieve the functional requirements of multi-moving state, a new autonomous underwater vehicle(AUV) provided with the functions such as the submarine vectorial thrust, landing on the sea bottom, wheel driving on the ground and crawling on the ground was designed. Then five new theories and methods were proposed about the motion mechanism of the AUV such as vectorial thruster technology, design of a new wheel propeller, kinematics and dynamics, navigation control and the ambient flow field in complex sea conditions, which can all conquer conventional technique shortages and predict the multi-moving state performance under wave disturbance. The theoretical research can realize the results such as a vectorial transmission shaft with the characteristics of spatial deflexion and continual circumgyratetion, parameterized design of the new wheel propeller with preferable open-water performance and intensity characteristics satisfying multi-moving state requirements, motion computation and kinetic analysis of AUV's arbitrary postures under wave disturbance, a second-order sliding mode controller with double-loop structure based on dynamic boundary layer that ensures AUV's trajectory high-precision tracking performance under wave disturbance, fast and exact prediction of the ambient flow field characteristics and the interaction mechanism between AUV hull and wheel propellers. The elaborate data obtained from the theoretical research can provide an important theoretical guidance and technical support for the manufacture of experimental prototype.展开更多
The present paper deals with very important practical problems of wide range of applications. The main target of the present paper is to track all moving boundaries that appear throughout the whole process when dealin...The present paper deals with very important practical problems of wide range of applications. The main target of the present paper is to track all moving boundaries that appear throughout the whole process when dealing with multi-moving boundary problems continuously with time up to the end of the process with high accuracy and minimum number of iterations. A new numerical iterative scheme based the boundary integral equation method is developed to track the moving boundaries as well as compute all unknowns in the problem. Three practical applications, one for vaporization and two for ablation were solved and their results were compared with finite element, heat balance integral and the source and sink results and a good agreement were obtained.展开更多
文摘目的生成式和传统人工智能模型是信息时代的关键工具。在这些技术的助力下,眼动过程的样本生成与识别显得尤为关键,它已成为深入研究认知机制的重要手段。为了推动生成式人工智能在眼动技术领域的应用发展,解决眼动样本生成及因网络深度增加而导致的不透明性和不可解释性问题,并深入挖掘与幼儿语言发展相关的眼动数据,方法采集4~6岁幼儿理解不同焦点结构的眼动数据,采用生成式人工智能模型-变分自编码器(variational autoencoder,VAE)和传统模型-多层感知器(multi-layer perceptron,MLP)识别眼动模式的发展差异并尝试生成新样本,基于灰色关联分析和混淆矩阵对生成式数据集进行解释。结果结果表明:(1)VAE生成的4岁组、5岁组和6岁组幼儿眼动数据集精度高于MINIST数据集(mixed National Institute of Standards and Technology database),且与MLP分析结果一致,具有准确性、多样性和一定的可解释性;(2)生成式眼动数据及混淆矩阵结果表明,在无焦点结构句式中,幼儿在4~5岁、5~6岁两个阶段理解水平均有提升,而宾语焦点结构和主语焦点结构的眼动特征在4~5岁变化较小,5~6岁变化较大,说明幼儿对焦点结构的理解在5岁是一个关键期,这符合幼儿焦点结构理解发展规律。结论提出的人工智能耦合分析方法,具备有效识别眼动特征发展模式的能力,并能据此生成可靠的新样本。这一方法不仅为生成式人工智能与眼动技术的融合开辟了新的途径,而且为复杂语言理解问题提供了全新的思考方向。
基金Project(51505491)supported by the National Natural Science Foundation of ChinaProject(ZR2014EEP019)supported by the Natural Science Foundation of Shandong Province,China
文摘In order to achieve the functional requirements of multi-moving state, a new autonomous underwater vehicle(AUV) provided with the functions such as the submarine vectorial thrust, landing on the sea bottom, wheel driving on the ground and crawling on the ground was designed. Then five new theories and methods were proposed about the motion mechanism of the AUV such as vectorial thruster technology, design of a new wheel propeller, kinematics and dynamics, navigation control and the ambient flow field in complex sea conditions, which can all conquer conventional technique shortages and predict the multi-moving state performance under wave disturbance. The theoretical research can realize the results such as a vectorial transmission shaft with the characteristics of spatial deflexion and continual circumgyratetion, parameterized design of the new wheel propeller with preferable open-water performance and intensity characteristics satisfying multi-moving state requirements, motion computation and kinetic analysis of AUV's arbitrary postures under wave disturbance, a second-order sliding mode controller with double-loop structure based on dynamic boundary layer that ensures AUV's trajectory high-precision tracking performance under wave disturbance, fast and exact prediction of the ambient flow field characteristics and the interaction mechanism between AUV hull and wheel propellers. The elaborate data obtained from the theoretical research can provide an important theoretical guidance and technical support for the manufacture of experimental prototype.
文摘The present paper deals with very important practical problems of wide range of applications. The main target of the present paper is to track all moving boundaries that appear throughout the whole process when dealing with multi-moving boundary problems continuously with time up to the end of the process with high accuracy and minimum number of iterations. A new numerical iterative scheme based the boundary integral equation method is developed to track the moving boundaries as well as compute all unknowns in the problem. Three practical applications, one for vaporization and two for ablation were solved and their results were compared with finite element, heat balance integral and the source and sink results and a good agreement were obtained.
