Neural Network Training for Average Calibration via Bregman Divergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks typically fail to achieve average calibration in predictive analysis, especially in survival analysis, leading to unreliable predictions and suboptimal decision-making in applications like medical treatments and equipment maintenance.
Innovation Solution
A method is developed to train neural networks by selecting subsets of feature and outcomes vectors, generating distribution vectors, producing prediction vectors, calculating Bregman divergence, and updating weights based on this divergence, which enables the network to satisfy average calibration and improve calibration performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If neural networks are used for predictive analysis, then modeling flexibility and complexity are improved, but average calibration performance deteriorates
Solution Approach 1:
The patent segments the training process into multiple iterations where each iteration focuses on a different subset of the data. By dividing the training data into subsets and training the neural network on each subset separately, the method maintains the flexibility of neural networks while improving calibration through focused, iterative optimization on smaller data portions.
Solution Approach 2:
The patent implements a feedback mechanism by calculating calibration metrics on validation subsets and using this information to guide further training iterations. The calibration performance is measured and fed back into the training process, allowing the neural network to adjust its parameters to improve average calibration while maintaining its predictive flexibility.
2Measurement precision
If simple prediction models are used, then average calibration is improved, but modeling capability deteriorates
Solution Approach 1:
The patent makes the training process dynamic by iteratively adjusting the neural network parameters based on calibration feedback from different data subsets. This dynamic approach allows the model to adapt its behavior to achieve better calibration while maintaining the inherent modeling capability of neural networks, effectively combining the strengths of both simple and complex models.
Data Source
AI summary
A method, which trains a neural network to perform an analysis that satisfies average calibration, includes a processor manipulating a data set that includes an outcomes vector and a set of feature vectors, each of which corresponds to one of the outcomes in the outcomes vector. The processor repeatedly: selects a subset of the set of feature vectors; generates a distribution vector for a subset of the outcomes vector that corresponds to the subset of the set of feature vectors; produces a prediction vector by running the neural network on the subset of the set of feature vectors; calculates a Bregman divergence between the distribution vector and a scoring distribution vector of the prediction vector; and updates weights of the neural network based on the Bregman divergence.