Prediction Value Correction Using Distribution Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Numerical prediction models face reduced performance when distances between selected similar examples and the current state are large, leading to deviations in prediction distributions and reduced accuracy.
Innovation Solution
An information processing apparatus that includes a processor for selecting similar cases, estimators for frequency data of observation and prediction values, and a corrector that adjusts prediction values based on cumulative distribution functions to align with observed data, enhancing prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If similar examples are selected from the database based on distance to current state, then prediction coverage is improved, but prediction accuracy deteriorates when distances are large
Solution Approach 1:
The system estimates the cumulative distribution function of prediction values from selected similar examples and compares it with the cumulative distribution function of observation values. Based on this feedback comparison, correction values are calculated and applied to adjust prediction values, thereby improving accuracy while maintaining broad prediction coverage
Solution Approach 2:
The system transforms prediction values by adding correction values derived from distribution function comparisons. This parameter change approach adjusts the prediction results to better match observed data distributions, resolving the accuracy issue while preserving the ability to handle diverse cases
2Device complexity
If a simple selection method is used to find similar cases, then device complexity is reduced, but prediction performance deteriorates when examples are far from current state
Solution Approach 1:
The cumulative distribution function acts as an intermediary between the simple selection process and the final prediction accuracy. By comparing distribution functions and using them to calculate correction values, the system bridges the gap between basic similarity selection and high-performance prediction without complicating the selection methodology
Data Source
AI summary
According to one embodiment, an information processing apparatus includes: a processor configured to select a first case based on subject data including at least one feature, and acquire a first prediction value that is a value of an objective variable included in the first case; a first estimator configured to estimate frequency data indicating frequencies of observation values of the objective variable, based on a history of observation values of the objective variable; a second estimator configured to estimate first frequency data indicating frequencies of first prediction values, based on a history of first prediction values acquired before the first prediction value is acquired; and a corrector configured to correct the first prediction value acquired by the processor, based on the frequency data and the first frequency data.