Intervention Effect Prediction via Distribution Bias Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting the effects of interventions, such as those in PTL 1, face challenges with selection bias and confounding bias, particularly in applying discrepancy distance to multiple interventions and achieving significant reduction in confounding bias.
Innovation Solution
A computer system employing machine learning models to map person state factors into a feature space, reducing distribution differences in training data, and using a second model to predict intervention effects, with a generative adversarial network (GAN) to adjust distribution biases and eliminate confounding influences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If discrepancy distance is used to adjust group distribution bias, then confounding bias adjustment is achieved, but it is difficult to apply to a plurality of interventions and the effect of reducing confounding bias is small
Solution Approach 1:
The patent segments the single discrepancy distance calculation into multiple distribution comparison operations, each handling a specific intervention type. The feature space is divided into multiple intervention-specific regions, allowing separate bias adjustment for each intervention while maintaining a unified feature representation framework.
Solution Approach 2:
The patent introduces an additional dimension to the feature space by incorporating intervention-type-specific features alongside the original features. This dimensional extension allows the model to simultaneously handle multiple intervention types while maintaining the ability to adjust distribution biases through the extended feature representation.
2Measurement precision
If machine learning models map training data to feature space, then prediction accuracy is improved, but selection bias and confounding bias remain unaddressed
Solution Approach 1:
The patent introduces distribution adjustment mechanisms as intermediary components between the feature mapping process and the prediction process. These intermediaries (distribution comparison modules and bias adjustment layers) mediate the transformation of features by explicitly correcting selection bias and confounding bias before final predictions are generated.
Solution Approach 2:
The patent applies preliminary anti-action by proactively identifying and correcting selection bias and confounding bias during the feature mapping stage, before these biases can affect the prediction outcomes. The distribution comparison and adjustment operations are performed in advance to counteract the harmful effects of bias in the training data.
Data Source
AI summary
A computer system manages a first model configured to generate a feature by mapping a vector including values of a plurality of factors representing a state of a person to a feature space and a second model configured to output predicted values of effects of a plurality of interventions on a person based on the feature, the first model and the second model being generated by machine learning. The first model maps a plurality of pieces of training data used in the machine learning to the feature space such that a difference in distribution of the plurality of pieces of training data in the feature space is reduced. The computer system receives input data including the values of the plurality of factors, generates the feature of the input data by inputting the input data into the first model, and calculates the predicted values of the effects of the plurality of interventions by inputting the feature of the input data into the second model.