The learning algorithms of causal discovery mainly include score-based methods and genetic algorithms(GA).The score-based algorithms are prone to searching space explosion.Classical GA is slow to converge,and prone to...The learning algorithms of causal discovery mainly include score-based methods and genetic algorithms(GA).The score-based algorithms are prone to searching space explosion.Classical GA is slow to converge,and prone to falling into local optima.To address these issues,an improved GA with domain knowledge(IGADK)is proposed.Firstly,domain knowledge is incorporated into the learning process of causality to construct a new fitness function.Secondly,a dynamical mutation operator is introduced in the algorithm to accelerate the convergence rate.Finally,an experiment is conducted on simulation data,which compares the classical GA with IGADK with domain knowledge of varying accuracy.The IGADK can greatly reduce the number of iterations,populations,and samples required for learning,which illustrates the efficiency and effectiveness of the proposed algorithm.展开更多
Causality,the science of cause and effect,has made it possible to create a new family of models.Such models are often referred to as causal models.Unlike those of mathematical,numerical,empirical,or machine learning(M...Causality,the science of cause and effect,has made it possible to create a new family of models.Such models are often referred to as causal models.Unlike those of mathematical,numerical,empirical,or machine learning(ML)nature,causal models hope to tie the cause(s)to the effect(s)pertaining to a phenomenon(i.e.,data generating process)through causal principles.This paper presents one of the first works at creating causal models in the area of structural and construction engineering.To this end,this paper starts with a brief review of the principles of causality and then adopts four causal discovery algorithms,namely,PC(Peter-Clark),FCI(fast causal inference),GES(greedy equivalence search),and GRa SP(greedy relaxation of the sparsest permutation),have been used to examine four phenomena,including predicting the load-bearing capacity of axially loaded members,fire resistance of structural members,shear strength of beams,and resistance of walls against impulsive(blast)loading.Findings from this study reveal the possibility and merit of discovering complete and partial causal models.Finally,this study also proposes two simple metrics that can help assess the performance of causal discovery algorithms.展开更多
Causal inference is a powerful modeling tool for explanatory analysis,which might enable current machine learning to become explainable.How to marry causal inference with machine learning to develop explainable artifi...Causal inference is a powerful modeling tool for explanatory analysis,which might enable current machine learning to become explainable.How to marry causal inference with machine learning to develop explainable artificial intelligence(XAI)algorithms is one of key steps toward to the artificial intelligence 2.0.With the aim of bringing knowledge of causal inference to scholars of machine learning and artificial intelligence,we invited researchers working on causal inference to write this survey from different aspects of causal inference.This survey includes the following sections:“Estimating average treatment effect:A brief review and beyond”from Dr.Kun Kuang,“Attribution problems in counterfactual inference”from Prof.Lian Li,“The Yule–Simpson paradox and the surrogate paradox”from Prof.Zhi Geng,“Causal potential theory”from Prof.Lei Xu,“Discovering causal information from observational data”from Prof.Kun Zhang,“Formal argumentation in causal reasoning and explanation”from Profs.Beishui Liao and Huaxin Huang,“Causal inference with complex experiments”from Prof.Peng Ding,“Instrumental variables and negative controls for observational studies”from Prof.Wang Miao,and“Causal inference with interference”from Dr.Zhichao Jiang.展开更多
Fault detection and diagnosis(FDD)plays a significant role in ensuring the safety and stability of chemical processes.With the development of artificial intelligence(AI)and big data technologies,data-driven approaches...Fault detection and diagnosis(FDD)plays a significant role in ensuring the safety and stability of chemical processes.With the development of artificial intelligence(AI)and big data technologies,data-driven approaches with excellent performance are widely used for FDD in chemical processes.However,improved predictive accuracy has often been achieved through increased model complexity,which turns models into black-box methods and causes uncertainty regarding their decisions.In this study,a causal temporal graph attention network(CTGAN)is proposed for fault diagnosis of chemical processes.A chemical causal graph is built by causal inference to represent the propagation path of faults.The attention mechanism and chemical causal graph were combined to help us notice the key variables relating to fault fluctuations.Experiments in the Tennessee Eastman(TE)process and the green ammonia(GA)process showed that CTGAN achieved high performance and good explainability.展开更多
Microblogging platforms like X(formerly Twitter)and Sina Weibo have become key channels for spreading information online.Accurately predicting information spread,such as users’reposting activities,is essential for ap...Microblogging platforms like X(formerly Twitter)and Sina Weibo have become key channels for spreading information online.Accurately predicting information spread,such as users’reposting activities,is essential for applications including content recommendation and analyzing public sentiment.Current advanced models rely on deep representation learning to extract features from various inputs,such as users’social connections and repost history,to forecast reposting behavior.Nonetheless,these models frequently ignore intrinsic confounding factors,which may cause the models to capture spurious relationships,ultimately impacting prediction performance.To address this limitation,we propose a novel Debiased Reposting Prediction model(DRP).Our model mitigates the influence of confounding variables by incorporating intervention operations from causal inference,enabling it to learn the causal associations between features and user reposting behavior.Specifically,we introduce a memory network within DRP to enhance the model’s perception of confounder distributions.This network aggregates and learns confounding information dispersed across different training data batches by optimizing the reconstruction loss.Furthermore,recognizing the challenge of acquiring prior knowledge of causal graphs,which is crucial for causal inference,we develop a causal discovery module within DRP(CD-DRP).This module allows the model to autonomously uncover the causal graph of feature variables by analyzing microblogging data.Experimental results on multiple real-world datasets demonstrate that our proposed method effectively uncovers causal relationships between variables,exhibits strong time efficiency,and outperforms state-of-the-art models in prediction performance(improved by 2.54%)and overfitting reduction(by 7.44%).展开更多
Gene regulatory network (GRN) inference from gene expression data is asignificant approach to understanding aspects of the biological system.Compared with generalized correlation-based methods, causality-inspiredones ...Gene regulatory network (GRN) inference from gene expression data is asignificant approach to understanding aspects of the biological system.Compared with generalized correlation-based methods, causality-inspiredones seem more rational to infer regulatory relationships. We proposeGRINCD, a novel GRN inference framework empowered by graph representationlearning and causal asymmetric learning, considering both linearand non-linear regulatory relationships. First, high-quality representation ofeach gene is generated using graph neural network. Then, we apply theadditive noise model to predict the causal regulation of each regulator-targetpair. Additionally, we design two channels and finally assemble them forrobust prediction. Through comprehensive comparisons of our frameworkwith state-of-the-art methods based on different principles on numerousdatasets of diverse types and scales, the experimental results show that ourframework achieves superior or comparable performance under variousevaluation metrics. Our work provides a new clue for constructing GRNs,and our proposed framework GRINCD also shows potential in identifyingkey factors affecting cancerdevelopment.展开更多
Transformer top oil temperature prediction is a research focal point in online monitoring of transformer operational status.Existing methods lack interpretability in feature selection and do not consider the temporal ...Transformer top oil temperature prediction is a research focal point in online monitoring of transformer operational status.Existing methods lack interpretability in feature selection and do not consider the temporal correlation of features.To address these issues,we provide a top oil temperature of electric transformers prediction method by using causal discovery and the Graph Neural Network(GNN)-Long Short-Term Memory(LSTM)model in this paper.To conduct feature selection,we use causal discovery to reduce feature dimensionality.Then,construct causal graph data based on the causal relationship matrix.Finally,spatiotemporal features are extracted by the GNN-LSTM model for oil temperature prediction.Experimental results demonstrate that this method can scientifically carry out feature selection,ensuring prediction accuracy and result robustness.展开更多
This research program enhances our understanding of how the Sun influences Earth,leading to cyclic global climate variations.The Sun exhibits many wellknown features and events that consistently impact Earth.These sol...This research program enhances our understanding of how the Sun influences Earth,leading to cyclic global climate variations.The Sun exhibits many wellknown features and events that consistently impact Earth.These solar phenomena directly affect the magnetosphere,which is the region of space around Earth shaped by its magnetic field.As a result of these interactions,changes occur in the atmospheric layers near the surface.Recognizing these links between solar activity,the magnetosphere,and atmospheric layers is essential for accurately understanding the processes behind climate variability and weather patterns.The influence of solar events on the lower atmosphere underscores the importance of integrating Sun-Earth interactions into climate research.Human impacts related to greenhouse gases are not entirely global;patterns of atmospheric and debris transport are mainly confined within the troposphere,stratosphere,and mesosphere,as well as by each hemisphere.Furthermore,these factors primarily restrict pollutants from circulating freely throughout Earth’s atmosphere.Human activities are adding organic and inorganic debris to land and oceans,altering the Earth’s crust.Densely populated areas are more heavily affected than less populated regions.Changes in atmospheric interactions,climate change,and global warming are likely to be more pronounced in densely populated regions and less so in areas with lower population density.The demand for vital services and resources—including food,sanitation,infrastructure,healthcare,housing,education,and leisure—is growing;consequently,emissions of various pollutants,such as methane and CO_(2),are expected to increase.The environmental degradation we observe today is worsened by various human-related factors,including intensive agriculture and livestock production,as well as the extraction of water and mineral resources.Human activities also lead to the release of plastics and other pollutants into the oceans,where ocean currents quickly spread materials that do not break down across the globe,causing persistent contamination because some substances biodegrade very slowly.This analysis is grounded in established principles of climate science and aims to be accessible across various disciplines,including environmental science,health,agriculture,economics,and policy.The conclusions highlight both immediate opportunities and long-term consequences that could emerge from the development of general-purpose intelligent systems,underscoring the importance of human-centered oversight,transparency,and equitable outcomes.展开更多
Model-based methods have recently been shown promising for offline reinforcement learning(RL),which aims at learning good policies from historical data without interacting with the environment.Previous model-based off...Model-based methods have recently been shown promising for offline reinforcement learning(RL),which aims at learning good policies from historical data without interacting with the environment.Previous model-based offline RL methods employ a straightforward prediction method that maps the states and actions directly to the next-step states.However,such a prediction method tends to capture spurious relations caused by the sampling policy preference behind the offline data.It is sensible that the environment model should focus on causal influences,which can facilitate learning an effective policy that can generalize well to unseen states.In this paper,we first provide theoretical results that causal environment models can outperform plain environment models in offline RL by incorporating the causal structure into the generalization error bound.We also propose a practical algorithm,oFfline mOdel-based reinforcement learning with CaUsal Structured World Models(FOCUS),to illustrate the feasibility of learning and leveraging causal structure in offline RL.Experimental results on two benchmarks show that FOCUS reconstructs the underlying causal structure accurately and robustly,and,as a result,outperforms both model-based offline RL algorithms and causal model-based offline RL algorithms.展开更多
The utilization of big Earth data has provided insights into the planet we inhabit in unprecedented dimensions and scales.Unraveling the concealed causal connections within intricate data holds paramount importance fo...The utilization of big Earth data has provided insights into the planet we inhabit in unprecedented dimensions and scales.Unraveling the concealed causal connections within intricate data holds paramount importance for attaining a profound comprehension of the Earth system.Statistical methods founded on correlation have predominated in Earth system science(ESS)for a long time.Nevertheless,correlation does not imply causation,especially when confronted with spurious correlations resulting from big data.Consequently,traditional correlation and regression methods are inadequate for addressing causation related problems in the Earth system.In recent years,propelled by advancements in causal theory and inference methods,particularly the maturity of causal discovery and causal graphical models,causal inference has demonstrated vigorous vitality in various research directions in the Earth system,such as regularities revealing,processes understanding,hypothesis testing,and physical models improving.This paper commences by delving into the origins,connotations,and development of causality,subsequently outlining the principal frameworks of causal inference and the commonly used methods in ESS.Additionally,it reviews the applications of causal inference in the main branches of the Earth system and summarizes the challenges and development directions of causal inference in ESS.In the big Earth data era,as an important method of big data analysis,causal inference,along with physical model and machine learning,can assist the paradigm transformation of ESS from a model-driven paradigm to a paradigm of integration of both mechanism and data.Looking forward,the establishment of a meticulously structured and normalized causal theory can act as a foundational cornerstone for fostering causal cognition in ESS and propel the leap from fragmented research towards a comprehensive understanding of the Earth system.展开更多
基金supported by the National Social Science Fund of China(2022-SKJJ-B-084).
文摘The learning algorithms of causal discovery mainly include score-based methods and genetic algorithms(GA).The score-based algorithms are prone to searching space explosion.Classical GA is slow to converge,and prone to falling into local optima.To address these issues,an improved GA with domain knowledge(IGADK)is proposed.Firstly,domain knowledge is incorporated into the learning process of causality to construct a new fitness function.Secondly,a dynamical mutation operator is introduced in the algorithm to accelerate the convergence rate.Finally,an experiment is conducted on simulation data,which compares the classical GA with IGADK with domain knowledge of varying accuracy.The IGADK can greatly reduce the number of iterations,populations,and samples required for learning,which illustrates the efficiency and effectiveness of the proposed algorithm.
文摘Causality,the science of cause and effect,has made it possible to create a new family of models.Such models are often referred to as causal models.Unlike those of mathematical,numerical,empirical,or machine learning(ML)nature,causal models hope to tie the cause(s)to the effect(s)pertaining to a phenomenon(i.e.,data generating process)through causal principles.This paper presents one of the first works at creating causal models in the area of structural and construction engineering.To this end,this paper starts with a brief review of the principles of causality and then adopts four causal discovery algorithms,namely,PC(Peter-Clark),FCI(fast causal inference),GES(greedy equivalence search),and GRa SP(greedy relaxation of the sparsest permutation),have been used to examine four phenomena,including predicting the load-bearing capacity of axially loaded members,fire resistance of structural members,shear strength of beams,and resistance of walls against impulsive(blast)loading.Findings from this study reveal the possibility and merit of discovering complete and partial causal models.Finally,this study also proposes two simple metrics that can help assess the performance of causal discovery algorithms.
文摘Causal inference is a powerful modeling tool for explanatory analysis,which might enable current machine learning to become explainable.How to marry causal inference with machine learning to develop explainable artificial intelligence(XAI)algorithms is one of key steps toward to the artificial intelligence 2.0.With the aim of bringing knowledge of causal inference to scholars of machine learning and artificial intelligence,we invited researchers working on causal inference to write this survey from different aspects of causal inference.This survey includes the following sections:“Estimating average treatment effect:A brief review and beyond”from Dr.Kun Kuang,“Attribution problems in counterfactual inference”from Prof.Lian Li,“The Yule–Simpson paradox and the surrogate paradox”from Prof.Zhi Geng,“Causal potential theory”from Prof.Lei Xu,“Discovering causal information from observational data”from Prof.Kun Zhang,“Formal argumentation in causal reasoning and explanation”from Profs.Beishui Liao and Huaxin Huang,“Causal inference with complex experiments”from Prof.Peng Ding,“Instrumental variables and negative controls for observational studies”from Prof.Wang Miao,and“Causal inference with interference”from Dr.Zhichao Jiang.
基金support of the National Key Research and Development Program of China(2021YFB4000505).
文摘Fault detection and diagnosis(FDD)plays a significant role in ensuring the safety and stability of chemical processes.With the development of artificial intelligence(AI)and big data technologies,data-driven approaches with excellent performance are widely used for FDD in chemical processes.However,improved predictive accuracy has often been achieved through increased model complexity,which turns models into black-box methods and causes uncertainty regarding their decisions.In this study,a causal temporal graph attention network(CTGAN)is proposed for fault diagnosis of chemical processes.A chemical causal graph is built by causal inference to represent the propagation path of faults.The attention mechanism and chemical causal graph were combined to help us notice the key variables relating to fault fluctuations.Experiments in the Tennessee Eastman(TE)process and the green ammonia(GA)process showed that CTGAN achieved high performance and good explainability.
文摘Microblogging platforms like X(formerly Twitter)and Sina Weibo have become key channels for spreading information online.Accurately predicting information spread,such as users’reposting activities,is essential for applications including content recommendation and analyzing public sentiment.Current advanced models rely on deep representation learning to extract features from various inputs,such as users’social connections and repost history,to forecast reposting behavior.Nonetheless,these models frequently ignore intrinsic confounding factors,which may cause the models to capture spurious relationships,ultimately impacting prediction performance.To address this limitation,we propose a novel Debiased Reposting Prediction model(DRP).Our model mitigates the influence of confounding variables by incorporating intervention operations from causal inference,enabling it to learn the causal associations between features and user reposting behavior.Specifically,we introduce a memory network within DRP to enhance the model’s perception of confounder distributions.This network aggregates and learns confounding information dispersed across different training data batches by optimizing the reconstruction loss.Furthermore,recognizing the challenge of acquiring prior knowledge of causal graphs,which is crucial for causal inference,we develop a causal discovery module within DRP(CD-DRP).This module allows the model to autonomously uncover the causal graph of feature variables by analyzing microblogging data.Experimental results on multiple real-world datasets demonstrate that our proposed method effectively uncovers causal relationships between variables,exhibits strong time efficiency,and outperforms state-of-the-art models in prediction performance(improved by 2.54%)and overfitting reduction(by 7.44%).
文摘Gene regulatory network (GRN) inference from gene expression data is asignificant approach to understanding aspects of the biological system.Compared with generalized correlation-based methods, causality-inspiredones seem more rational to infer regulatory relationships. We proposeGRINCD, a novel GRN inference framework empowered by graph representationlearning and causal asymmetric learning, considering both linearand non-linear regulatory relationships. First, high-quality representation ofeach gene is generated using graph neural network. Then, we apply theadditive noise model to predict the causal regulation of each regulator-targetpair. Additionally, we design two channels and finally assemble them forrobust prediction. Through comprehensive comparisons of our frameworkwith state-of-the-art methods based on different principles on numerousdatasets of diverse types and scales, the experimental results show that ourframework achieves superior or comparable performance under variousevaluation metrics. Our work provides a new clue for constructing GRNs,and our proposed framework GRINCD also shows potential in identifyingkey factors affecting cancerdevelopment.
文摘Transformer top oil temperature prediction is a research focal point in online monitoring of transformer operational status.Existing methods lack interpretability in feature selection and do not consider the temporal correlation of features.To address these issues,we provide a top oil temperature of electric transformers prediction method by using causal discovery and the Graph Neural Network(GNN)-Long Short-Term Memory(LSTM)model in this paper.To conduct feature selection,we use causal discovery to reduce feature dimensionality.Then,construct causal graph data based on the causal relationship matrix.Finally,spatiotemporal features are extracted by the GNN-LSTM model for oil temperature prediction.Experimental results demonstrate that this method can scientifically carry out feature selection,ensuring prediction accuracy and result robustness.
文摘This research program enhances our understanding of how the Sun influences Earth,leading to cyclic global climate variations.The Sun exhibits many wellknown features and events that consistently impact Earth.These solar phenomena directly affect the magnetosphere,which is the region of space around Earth shaped by its magnetic field.As a result of these interactions,changes occur in the atmospheric layers near the surface.Recognizing these links between solar activity,the magnetosphere,and atmospheric layers is essential for accurately understanding the processes behind climate variability and weather patterns.The influence of solar events on the lower atmosphere underscores the importance of integrating Sun-Earth interactions into climate research.Human impacts related to greenhouse gases are not entirely global;patterns of atmospheric and debris transport are mainly confined within the troposphere,stratosphere,and mesosphere,as well as by each hemisphere.Furthermore,these factors primarily restrict pollutants from circulating freely throughout Earth’s atmosphere.Human activities are adding organic and inorganic debris to land and oceans,altering the Earth’s crust.Densely populated areas are more heavily affected than less populated regions.Changes in atmospheric interactions,climate change,and global warming are likely to be more pronounced in densely populated regions and less so in areas with lower population density.The demand for vital services and resources—including food,sanitation,infrastructure,healthcare,housing,education,and leisure—is growing;consequently,emissions of various pollutants,such as methane and CO_(2),are expected to increase.The environmental degradation we observe today is worsened by various human-related factors,including intensive agriculture and livestock production,as well as the extraction of water and mineral resources.Human activities also lead to the release of plastics and other pollutants into the oceans,where ocean currents quickly spread materials that do not break down across the globe,causing persistent contamination because some substances biodegrade very slowly.This analysis is grounded in established principles of climate science and aims to be accessible across various disciplines,including environmental science,health,agriculture,economics,and policy.The conclusions highlight both immediate opportunities and long-term consequences that could emerge from the development of general-purpose intelligent systems,underscoring the importance of human-centered oversight,transparency,and equitable outcomes.
文摘Model-based methods have recently been shown promising for offline reinforcement learning(RL),which aims at learning good policies from historical data without interacting with the environment.Previous model-based offline RL methods employ a straightforward prediction method that maps the states and actions directly to the next-step states.However,such a prediction method tends to capture spurious relations caused by the sampling policy preference behind the offline data.It is sensible that the environment model should focus on causal influences,which can facilitate learning an effective policy that can generalize well to unseen states.In this paper,we first provide theoretical results that causal environment models can outperform plain environment models in offline RL by incorporating the causal structure into the generalization error bound.We also propose a practical algorithm,oFfline mOdel-based reinforcement learning with CaUsal Structured World Models(FOCUS),to illustrate the feasibility of learning and leveraging causal structure in offline RL.Experimental results on two benchmarks show that FOCUS reconstructs the underlying causal structure accurately and robustly,and,as a result,outperforms both model-based offline RL algorithms and causal model-based offline RL algorithms.
基金supported by the Basic Science Center for Tibetan Plateau Earth System(BCTPES,NSFC project Grant Nos.41988101)the National Natural Science Foundation of China(Grant No.42101397)。
文摘The utilization of big Earth data has provided insights into the planet we inhabit in unprecedented dimensions and scales.Unraveling the concealed causal connections within intricate data holds paramount importance for attaining a profound comprehension of the Earth system.Statistical methods founded on correlation have predominated in Earth system science(ESS)for a long time.Nevertheless,correlation does not imply causation,especially when confronted with spurious correlations resulting from big data.Consequently,traditional correlation and regression methods are inadequate for addressing causation related problems in the Earth system.In recent years,propelled by advancements in causal theory and inference methods,particularly the maturity of causal discovery and causal graphical models,causal inference has demonstrated vigorous vitality in various research directions in the Earth system,such as regularities revealing,processes understanding,hypothesis testing,and physical models improving.This paper commences by delving into the origins,connotations,and development of causality,subsequently outlining the principal frameworks of causal inference and the commonly used methods in ESS.Additionally,it reviews the applications of causal inference in the main branches of the Earth system and summarizes the challenges and development directions of causal inference in ESS.In the big Earth data era,as an important method of big data analysis,causal inference,along with physical model and machine learning,can assist the paradigm transformation of ESS from a model-driven paradigm to a paradigm of integration of both mechanism and data.Looking forward,the establishment of a meticulously structured and normalized causal theory can act as a foundational cornerstone for fostering causal cognition in ESS and propel the leap from fragmented research towards a comprehensive understanding of the Earth system.
