Deep Learning Architecture for Incidental Truncation Bias
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for addressing incidental truncation bias are not adaptable to deep learning models, which are commonly used for predicting multiple-sequential-event based outcomes, leading to biased predictions due to non-random sample selection.
Innovation Solution
The implementation of deep learning architectures that utilize sequential convolutional neural networks (CNNs) to predict multiple-sequential-event based outcomes by selectively feeding input variables to internal layers of neural networks, accounting for incidental truncation through a two-stage prediction model where the output of a penultimate layer is combined with additional input variables to reduce bias.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional statistical techniques (e.g., Heckman two-stage modelling) are used to account for incidental truncation bias, then measurement precision can be improved, but adaptability to deep learning models deteriorates
Solution Approach 1:
The patent changes the fundamental parameters of the modeling approach by transitioning from conventional statistical methods to deep learning architectures. Specifically, it modifies the input processing parameters by selectively feeding different subsets of variables to different neural networks, and changes the computational parameters by using sequential CNNs with specific layer configurations (input layers, internal layers with appended variables, output layers). This enables the system to handle incidental truncation bias while maintaining adaptability to deep learning frameworks.
2Measurement precision
If all variables are fed to every layer of the neural network, then device complexity is reduced, but measurement precision deteriorates due to incidental truncation bias
Solution Approach 1:
The patent segments the variable input across different neural network layers and different networks. Specifically, it divides the plurality of variables into first input variables and second input variables, where not all variables are fed to every layer. The first deep neural network receives first input variables in its input layer and additional second input variables in its internal layer, while the second deep neural network receives different variable combinations. This segmentation allows the model to account for incidental truncation bias by selectively processing variables at different stages, improving prediction accuracy without requiring every variable to be processed throughout the entire network.
Solution Approach 2:
The patent applies partial action by selectively feeding only specific subsets of variables to specific layers rather than all variables to all layers. The first input layer receives a first subset of variables, and internal layers receive additional second subset variables appended to previous layer outputs. This partial processing of variables at different stages enables the model to handle the complexity of sequential events and truncation bias without the excessive computational burden of processing all variables at every layer.
3Measurement precision
If separate models are created for each decision in a multiple-sequential-event outcome, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent merges multiple decision predictions into a unified deep learning architecture. Instead of creating completely separate models for each decision, it integrates them through a coordinated system where the first deep neural network predicts a first decision and the second deep neural network predicts a second decision, with the networks sharing computational resources and data processing infrastructure. The networks are trained using the same training data and share similar architectural components, reducing overall system complexity while maintaining the ability to accurately predict multiple sequential decisions.
Solution Approach 2:
The patent creates a universal deep learning architecture that can handle multiple decision predictions within a single integrated framework. Both the first and second deep neural networks use the same input processing mechanisms, variable selection approaches, and training procedures, making the system multi-functional for predicting different decisions in sequential events. This universal approach allows the model to handle various decision types using the same architectural patterns, reducing the need for entirely separate specialized models.
Data Source
AI summary
A system for predicting multiple-sequential-event based outcomes may include: a first deep neural network configured to predict a first decision, and that includes: a first input layer configured to receive as input less than an entirety of a plurality of variables; and an internal layer configured to receive as input a remainder of the plurality of variables appended to an output of a preceding layer, such that the first deep neural network is configured generate a prediction value for the first decision; and a second deep neural network configured to predict a second decision subsequent to the first decision, and that includes: a second input layer configured to receive as input a representation of output from a penultimate layer appended to at least a portion of the plurality of variables, wherein the second deep neural network is configured to generate a prediction value for the second decision.