Causal Inference and Learning

The field of causal inference and learning is moving towards developing more robust and innovative methods for modeling complex causal relationships and interactions. One notable direction is the integration of causal inference with deep learning techniques, such as transformers and reinforcement learning, to improve the accuracy and interpretability of causal models. Another significant trend is the development of methods for incremental causal graph learning, which enables the detection of evolving causal relationships in real-time settings. Additionally, researchers are exploring the use of positive-unlabeled learning for control group construction in observational causal inference, as well as the application of causal inference to imitation learning and agent decision-making. Noteworthy papers include CaSTFormer, which proposes a novel causal spatio-temporal transformer for driving intention prediction, and Reframing attention as a reinforcement learning problem for causal discovery, which introduces a causal process framework for representing dynamic hypotheses about causal structure. Other notable works include Incremental Causal Graph Learning for Online Cyberattack Detection and Positive-Unlabeled Learning for Control Group Construction in Observational Causal Inference.

Sources

CaSTFormer: Causal Spatio-Temporal Transformer for Driving Intention Prediction

Reframing attention as a reinforcement learning problem for causal discovery

Incremental Causal Graph Learning for Online Cyberattack Detection in Cyber-Physical Infrastructures

Positive-Unlabeled Learning for Control Group Construction in Observational Causal Inference

From Kicking to Causality: Simulating Infant Agency Detection with a Robust Intrinsic Reward

Identifying Conditional Causal Effects in MPDAGs

Confounded Causal Imitation Learning with Instrumental Variables

Built with on top of