UTS | causal reasoning random intervention based on Reinforcement Learning

【 title 】Stochastic Intervention for Causal Inference via Reinforcement Learning

【 The author team 】Tri Dung Duong, Qian Li, Guandong Xu

【 Date of publication 】2022.1.29

【 Thesis link 】https://www.sciencedirect.com/science/article/pii/S0925231222001072#!

【 Recommended reasons 】 Causal reasoning is widely used in precision medicine 、 Various decision-making fields such as optimal strategy and Economics . The main focus of causal reasoning is the treatment effect estimation of intervention strategies , Most of the existing methods are limited to deterministic treatment , The results of different treatments were compared . However, they lack the ability to solve fine-grained treatment effect estimation to enhance decision-making application . This paper proposes a new effective framework to estimate the treatment effect under random intervention to promote the research of causal reasoning . This paper develops a random intervention effect estimator based on nonparametric influence function （SIE）, Theoretical guarantee of robustness and fast convergence speed . Besides , This paper constructs a customized reinforcement learning algorithm based on random search solver , Can effectively find the optimal strategy , So as to produce the maximum expected results for the decision-making process . Last , This paper has conducted extensive empirical experiments , To verify that this framework can achieve excellent performance compared with the most advanced baseline .