This paper addresses the semi-parametric problem of estimating a low-dimensional parameter \( \theta_0 \) in the presence of high-dimensional nuisance parameters \( \eta_0 \). The authors propose a method called Double/Debiased Machine Learning (DML) to address the issue of regularization bias and overfitting, which can lead to substantial bias in the estimation of \( \theta_0 \). DML involves two key steps: (1) using Neyman-orthogonal moments/scores that are less sensitive to nuisance parameters, and (2) employing cross-fitting to efficiently split the data. These steps help remove the impact of regularization bias and overfitting, ensuring that the estimator of \( \theta_0 \) is approximately unbiased and normally distributed. The paper provides theoretical foundations for DML, demonstrating its validity under weak conditions and showing its effectiveness in various empirical applications, including partially linear regression models, instrumental variables models, and average treatment effect analysis.This paper addresses the semi-parametric problem of estimating a low-dimensional parameter \( \theta_0 \) in the presence of high-dimensional nuisance parameters \( \eta_0 \). The authors propose a method called Double/Debiased Machine Learning (DML) to address the issue of regularization bias and overfitting, which can lead to substantial bias in the estimation of \( \theta_0 \). DML involves two key steps: (1) using Neyman-orthogonal moments/scores that are less sensitive to nuisance parameters, and (2) employing cross-fitting to efficiently split the data. These steps help remove the impact of regularization bias and overfitting, ensuring that the estimator of \( \theta_0 \) is approximately unbiased and normally distributed. The paper provides theoretical foundations for DML, demonstrating its validity under weak conditions and showing its effectiveness in various empirical applications, including partially linear regression models, instrumental variables models, and average treatment effect analysis.