Keynote speech 1
Title: Learning Causality and Using Causality for Learning
Presenter: Prof. Kun Zhang, Carnegie Mellon University
Abstract: Can we find the causal direction between two random variables without temporal precedence information? Can we figure out where latent causal variables should be and how they are related? In our daily life and science, people often attempt to answer such causal questions for the purpose of understanding, proper manipulation of systems, and robust prediction under interventions. Moreover, we are concerned with issues with artificial intelligence (AI) in complex environments. For instance, how can we do transfer learning in a principled way? How can machines deal with adversarial attacks? Interestingly, it has recently been shown that causal information can facilitate understanding and solving various AI problems. This talk focused on how to learn causal representations (with or without latent variables) from observational data and why and how the causal perspective allows adaptive prediction and a potentially higher level of artificial intelligence.
Biography: Kun Zhang is an associate professor of philosophy and an affiliate faculty in the machine learning department of Carnegie Mellon University. He has been actively developing methods for automated causal discovery from various kinds of data and investigating machine learning problems including transfer learning and representation learning from a causal perspective. He has been frequently serving as a senior area chair, area chair, or senior program committee member for conferences in machine learning or artificial intelligence, including NeurIPS, ICML, UAI, IJCAI, AISTATS, and ICLR, and has co-organized a number of conferences or workshops to foster interdisciplinary exploration of causality.
Keynote speech 2
Title: LiNGAM approach to causal discovery
Presenter: Prof. Shohei Shimizu, Shiga University and RIKEN
Abstract: Causal discovery is a statistical methodology for inferring causal graphs that represent the causal structures of variables. Such causal graphs are the key to perform causal analysis. Causal structural information contained in causal graphs enables researchers to find the right sets of variables to be adjusted for (if any) for estimating causal effects, decompose them into direct and indirect causal effects, and better understand the causal mechanism, for example. This kind of causal analysis has been more often used in machine learning, e.g., for constructing interpretable and fair predictive models. A common way to draw causal graphs is to use domain knowledge. However, in many cases, enough domain knowledge is not available. Thus, in such cases, data-driven causal discovery methods help researchers draw causal graphs.
In this talk, I first briefly review causal discovery methods and their applications. Then, I introduce our recent two works on causal discovery in the presence of hidden variables. These works base the idea of linear non-Gaussian acyclic models, LiNGAM. The first proposes to estimate causal directions of observed variables that have no unobserved common causes after detecting observed variables between which unobserved variables exist. The other aims to find the causal structure of latent factors from multiple datasets obtained under different conditions. It further analyzes which latent factors are common to the conditions and which are specific to some conditions.
Biography: Shohei Shimizu is a Professor at the Faculty of Data Science, Shiga University, Japan and leads the Causal Inference Team, RIKEN Center for Advanced Intelligence Project. He received a Ph.D. in Engineering from Osaka University in 2006. His research interests include statistical methodologies for learning causal relationships and their applications. He received Hayashi Chikio Award (Excellence Award) from the Behaviormetric Society in 2016. He serves as a coordinating editor of Behaviormetrika and as an associate editor of Neurocomputing and Neural Networks.
Title: Learning Causality and Using Causality for Learning
Presenter: Prof. Kun Zhang, Carnegie Mellon University
Abstract: Can we find the causal direction between two random variables without temporal precedence information? Can we figure out where latent causal variables should be and how they are related? In our daily life and science, people often attempt to answer such causal questions for the purpose of understanding, proper manipulation of systems, and robust prediction under interventions. Moreover, we are concerned with issues with artificial intelligence (AI) in complex environments. For instance, how can we do transfer learning in a principled way? How can machines deal with adversarial attacks? Interestingly, it has recently been shown that causal information can facilitate understanding and solving various AI problems. This talk focused on how to learn causal representations (with or without latent variables) from observational data and why and how the causal perspective allows adaptive prediction and a potentially higher level of artificial intelligence.
Biography: Kun Zhang is an associate professor of philosophy and an affiliate faculty in the machine learning department of Carnegie Mellon University. He has been actively developing methods for automated causal discovery from various kinds of data and investigating machine learning problems including transfer learning and representation learning from a causal perspective. He has been frequently serving as a senior area chair, area chair, or senior program committee member for conferences in machine learning or artificial intelligence, including NeurIPS, ICML, UAI, IJCAI, AISTATS, and ICLR, and has co-organized a number of conferences or workshops to foster interdisciplinary exploration of causality.
Keynote speech 2
Title: LiNGAM approach to causal discovery
Presenter: Prof. Shohei Shimizu, Shiga University and RIKEN
Abstract: Causal discovery is a statistical methodology for inferring causal graphs that represent the causal structures of variables. Such causal graphs are the key to perform causal analysis. Causal structural information contained in causal graphs enables researchers to find the right sets of variables to be adjusted for (if any) for estimating causal effects, decompose them into direct and indirect causal effects, and better understand the causal mechanism, for example. This kind of causal analysis has been more often used in machine learning, e.g., for constructing interpretable and fair predictive models. A common way to draw causal graphs is to use domain knowledge. However, in many cases, enough domain knowledge is not available. Thus, in such cases, data-driven causal discovery methods help researchers draw causal graphs.
In this talk, I first briefly review causal discovery methods and their applications. Then, I introduce our recent two works on causal discovery in the presence of hidden variables. These works base the idea of linear non-Gaussian acyclic models, LiNGAM. The first proposes to estimate causal directions of observed variables that have no unobserved common causes after detecting observed variables between which unobserved variables exist. The other aims to find the causal structure of latent factors from multiple datasets obtained under different conditions. It further analyzes which latent factors are common to the conditions and which are specific to some conditions.
Biography: Shohei Shimizu is a Professor at the Faculty of Data Science, Shiga University, Japan and leads the Causal Inference Team, RIKEN Center for Advanced Intelligence Project. He received a Ph.D. in Engineering from Osaka University in 2006. His research interests include statistical methodologies for learning causal relationships and their applications. He received Hayashi Chikio Award (Excellence Award) from the Behaviormetric Society in 2016. He serves as a coordinating editor of Behaviormetrika and as an associate editor of Neurocomputing and Neural Networks.