Tool variable generation and counter-fact reasoning method and device based on neural network

Abstract

The invention discloses a tool variable generation and counter-fact reasoning method and device based on a neural network. Aiming at the problem that a previous tool variable-based counter factual reasoning (such as handwritten numeric recognition) method needs to define and acquire tool variables in advance, the tool variables are directly learned and decoupled from observable variables, so that the causal inference efficiency is greatly improved, and the time and the cost are saved. According to the method, the tool variables are automatically extracted from the observable variables for the first time, and the method has originality and uniqueness in algorithm and application. The method is applied to an existing tool variable-based counter-fact prediction method, and compared with a method using a real tool variable, the performance causal inference is obviously improved. The method focuses on decoupling the representation of the tool variables from the observable variables, solves the problem that the tool variable-based counter-fact prediction technology needs to use prior knowledge and high cost to obtain tool variable data in advance, and improves the precision in the fields of handwritten numeral recognition and the like.

Description

technical field [0001] The present invention relates to the field of causal inference, in particular to an automatic instrumental variable decoupling method, which realizes a counterfactual prediction method that can directly extract instrumental variables from observable variables, thereby improving the efficiency and accuracy of handwritten digit recognition. Background technique [0002] Causal inference is dedicated to estimating the counterfactual results of interventions and assisting decision makers in making choices to achieve the goal of optimizing the results. The gold standard for causal inference is to use randomized controlled trials to randomly assign intervention values ​​for causal inference, but such methods are cost-prohibitive or even impossible. Some methods use weighting and matching methods to constrain the confounding variables that affect causal inference, but such methods can only be used when the confounding is completely observable, and when the co...

AI Technical Summary

Problems solved by technology

The current causal inference methods based on instrumental variables all require a pre-defined instrumental variable, but this is often not practical in real situations

Images