Action model learning has become a hot topic in knowledge engineering for automated planning.A key problem for learning action models is to analyze state changes before and after action executions from observed"p...Action model learning has become a hot topic in knowledge engineering for automated planning.A key problem for learning action models is to analyze state changes before and after action executions from observed"plan traces".To support such an analysis,a new approach is proposed to partition propositions of plan traces into states.First,vector representations of propositions and actions are obtained by training a neural network called Skip-Gram borrowed from the area of natural language processing(NLP).Then,a type of semantic distance among propositions and actions is defined based on their similarity measures in the vector space.Finally,k-means and k-nearest neighbor(kNN)algorithms are exploited to map propositions to states.This approach is called state partition by word vector(SPWV),which is implemented on top of a recent action model learning framework by Rao et al.Experimental results on the benchmark domains show that SPWV leads to a lower error rate of the learnt action model,compared to the probability based approach for state partition that was developed by Rao et al.展开更多
This paper proposes a method to recognize human-object interactions by modeling context between human actions and interacted objects.Human-object interaction recognition is a challenging task due to severe occlusion b...This paper proposes a method to recognize human-object interactions by modeling context between human actions and interacted objects.Human-object interaction recognition is a challenging task due to severe occlusion between human and objects during the interacting process.Since that human actions and interacted objects provide strong context information,i.e.some actions are usually related to some specific objects,the accuracy of recognition is significantly improved for both of them.Through the proposed method,both global and local temporal features from skeleton sequences are extracted to model human actions.In the meantime,kernel features are utilized to describe interacted objects.Finally,all possible solutions from actions and objects are optimized by modeling the context between them.The results of experiments demonstrate the effectiveness of our method.展开更多
With the passage of time, it has become important to investigate new methods for updating data to better fit the trends of the grey prediction model. The traditional GM（1,1） usually sets the grey action quantity as ...With the passage of time, it has become important to investigate new methods for updating data to better fit the trends of the grey prediction model. The traditional GM（1,1） usually sets the grey action quantity as a constant. Therefore, it cannot effectively fit the dynamic characteristics of the sequence, which results in the grey model having a low precision. The linear grey action quantity model cannot represent the index change law. This paper presents a grey action quantity model, the exponential optimization grey model（EOGM（1,1））, based on the exponential type of grey action quantity; it is constructed based on the exponential characteristics of the grey prediction model. The model can fully reflect the exponential characteristics of the simulation series with time. The exponential sequence has a higher fitting accuracy. The optimized result is verified using a numerical example for the fluctuating sequence and a case study for the index of the tertiary industry＇s GDP. The results show that the model improves the precision of the grey forecasting model and reduces the prediction error.展开更多
In order to build the model of the drum level wave action and sloshing, based on the method of modularization modeling, the hydrodynamic model of drum level wave action and sloshing was developed, and dynamic simulati...In order to build the model of the drum level wave action and sloshing, based on the method of modularization modeling, the hydrodynamic model of drum level wave action and sloshing was developed, and dynamic simulation researches were carried out based on the model. The results indicate that both drum level and drum length have functional relations with period of drum level wave action and sloshing. When the drum level decreases or drum length increases, the period of drum level wave action and sloshing increases, density of liquid and number of sub-module division have little influence on the period of drum level wave action and sloshing. The model was validated by the analytical solution theory of liquid’s wave action and sloshing in cuboid container, and the 3D graphics of drum level wave action and sloshing was also obtained. The model can dynamically reflect the rules of wave action and sloshing of water in the container exactly.展开更多
Action recognition is an important topic in computer vision. Recently, deep learning technologies have been successfully used in lots of applications including video data for sloving recognition problems. However, mos...Action recognition is an important topic in computer vision. Recently, deep learning technologies have been successfully used in lots of applications including video data for sloving recognition problems. However, most existing deep learning based recognition frameworks are not optimized for action in the surveillance videos. In this paper, we propose a novel method to deal with the recognition of different types of actions in outdoor surveillance videos. The proposed method first introduces motion compensation to improve the detection of human target. Then, it uses three different types of deep models with single and sequenced images as inputs for the recognition of different types of actions. Finally, predictions from different models are fused with a linear model. Experimental results show that the proposed method works well on the real surveillance videos.展开更多
In this paper, the numerical modelling of the tidal level and current in the Bohai Sea was carried out with ADI method, by taking the sum of four main tidal components M2,S2K2,O1 as the open boundary condition. The ca...In this paper, the numerical modelling of the tidal level and current in the Bohai Sea was carried out with ADI method, by taking the sum of four main tidal components M2,S2K2,O1 as the open boundary condition. The calculated values were consistent with the predicted ones (the observed values in the case of calm) in the Tidal Table. On the basis of the modelling of the tide, the sea level and current fields under the effects of strong wind were simulated. The calculated results were also quite satisfactory.展开更多
预训练世界模型是提升强化学习样本效率的关键技术,但现有方法因视频数据缺乏显式动作标注,难以捕捉状态转移的因果机制。对此,提出多模态大模型辅助的视频动作生成预训练框架(MLM-generated Action-based Pre-training from videos for...预训练世界模型是提升强化学习样本效率的关键技术,但现有方法因视频数据缺乏显式动作标注,难以捕捉状态转移的因果机制。对此,提出多模态大模型辅助的视频动作生成预训练框架(MLM-generated Action-based Pre-training from videos for world models,MAPO),通过整合视觉语言模型的语义理解能力与动力学建模需求,突破传统预训练范式在动作语义缺失方面的局限性。具体地,MAPO在预训练阶段利用多模态大模型(QWEN2_5-VL-7B)解析视频帧序列,生成细粒度语义动作描述,构建具有因果解释性的动作-状态关联;设计上下文量化编码机制,解耦场景静态特征与动态控制因素,增强跨模态表征能力。在微调阶段,通过双网络协同架构实现预训练动力学特征与真实环境动作的端到端对齐。实验表明,MAPO在DeepMind Control Suite和Meta-World的8项任务中的平均回报较最优基线获得稳定提升,尤其在长时程任务中展现出卓越的性能。该研究为跨模态世界模型训练提供了新范式,揭示了语义动作生成在因果推理中的关键作用。展开更多
基金Supported by the National Natural Science Foundation of China(61103136,61370156,61503074)Open Research Foundation of Science and Technology on Aerospace Flight Dynamics Laboratory(2014afdl002)
文摘Action model learning has become a hot topic in knowledge engineering for automated planning.A key problem for learning action models is to analyze state changes before and after action executions from observed"plan traces".To support such an analysis,a new approach is proposed to partition propositions of plan traces into states.First,vector representations of propositions and actions are obtained by training a neural network called Skip-Gram borrowed from the area of natural language processing(NLP).Then,a type of semantic distance among propositions and actions is defined based on their similarity measures in the vector space.Finally,k-means and k-nearest neighbor(kNN)algorithms are exploited to map propositions to states.This approach is called state partition by word vector(SPWV),which is implemented on top of a recent action model learning framework by Rao et al.Experimental results on the benchmark domains show that SPWV leads to a lower error rate of the learnt action model,compared to the probability based approach for state partition that was developed by Rao et al.
文摘This paper proposes a method to recognize human-object interactions by modeling context between human actions and interacted objects.Human-object interaction recognition is a challenging task due to severe occlusion between human and objects during the interacting process.Since that human actions and interacted objects provide strong context information,i.e.some actions are usually related to some specific objects,the accuracy of recognition is significantly improved for both of them.Through the proposed method,both global and local temporal features from skeleton sequences are extracted to model human actions.In the meantime,kernel features are utilized to describe interacted objects.Finally,all possible solutions from actions and objects are optimized by modeling the context between them.The results of experiments demonstrate the effectiveness of our method.
基金supported by the National Key Research and Development Program of China(2016YFC1402000)the National Science Foundation of China(41701593+2 种基金7137109871571157)the National Social Science Fund Major Project(14ZDB151)
文摘With the passage of time, it has become important to investigate new methods for updating data to better fit the trends of the grey prediction model. The traditional GM（1,1） usually sets the grey action quantity as a constant. Therefore, it cannot effectively fit the dynamic characteristics of the sequence, which results in the grey model having a low precision. The linear grey action quantity model cannot represent the index change law. This paper presents a grey action quantity model, the exponential optimization grey model（EOGM（1,1））, based on the exponential type of grey action quantity; it is constructed based on the exponential characteristics of the grey prediction model. The model can fully reflect the exponential characteristics of the simulation series with time. The exponential sequence has a higher fitting accuracy. The optimized result is verified using a numerical example for the fluctuating sequence and a case study for the index of the tertiary industry＇s GDP. The results show that the model improves the precision of the grey forecasting model and reduces the prediction error.
基金Project(200310) supported by Edison Research Foundation from General Electric (GE) in USAProject(59976022) supported by the National Natural Science Foundation of China
文摘In order to build the model of the drum level wave action and sloshing, based on the method of modularization modeling, the hydrodynamic model of drum level wave action and sloshing was developed, and dynamic simulation researches were carried out based on the model. The results indicate that both drum level and drum length have functional relations with period of drum level wave action and sloshing. When the drum level decreases or drum length increases, the period of drum level wave action and sloshing increases, density of liquid and number of sub-module division have little influence on the period of drum level wave action and sloshing. The model was validated by the analytical solution theory of liquid’s wave action and sloshing in cuboid container, and the 3D graphics of drum level wave action and sloshing was also obtained. The model can dynamically reflect the rules of wave action and sloshing of water in the container exactly.
文摘Action recognition is an important topic in computer vision. Recently, deep learning technologies have been successfully used in lots of applications including video data for sloving recognition problems. However, most existing deep learning based recognition frameworks are not optimized for action in the surveillance videos. In this paper, we propose a novel method to deal with the recognition of different types of actions in outdoor surveillance videos. The proposed method first introduces motion compensation to improve the detection of human target. Then, it uses three different types of deep models with single and sequenced images as inputs for the recognition of different types of actions. Finally, predictions from different models are fused with a linear model. Experimental results show that the proposed method works well on the real surveillance videos.
文摘In this paper, the numerical modelling of the tidal level and current in the Bohai Sea was carried out with ADI method, by taking the sum of four main tidal components M2,S2K2,O1 as the open boundary condition. The calculated values were consistent with the predicted ones (the observed values in the case of calm) in the Tidal Table. On the basis of the modelling of the tide, the sea level and current fields under the effects of strong wind were simulated. The calculated results were also quite satisfactory.
文摘预训练世界模型是提升强化学习样本效率的关键技术,但现有方法因视频数据缺乏显式动作标注,难以捕捉状态转移的因果机制。对此,提出多模态大模型辅助的视频动作生成预训练框架(MLM-generated Action-based Pre-training from videos for world models,MAPO),通过整合视觉语言模型的语义理解能力与动力学建模需求,突破传统预训练范式在动作语义缺失方面的局限性。具体地,MAPO在预训练阶段利用多模态大模型(QWEN2_5-VL-7B)解析视频帧序列,生成细粒度语义动作描述,构建具有因果解释性的动作-状态关联;设计上下文量化编码机制,解耦场景静态特征与动态控制因素,增强跨模态表征能力。在微调阶段,通过双网络协同架构实现预训练动力学特征与真实环境动作的端到端对齐。实验表明,MAPO在DeepMind Control Suite和Meta-World的8项任务中的平均回报较最优基线获得稳定提升,尤其在长时程任务中展现出卓越的性能。该研究为跨模态世界模型训练提供了新范式,揭示了语义动作生成在因果推理中的关键作用。
