Rule Learning by Sequential Weight Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing Boosting algorithms, particularly AdaBoost, face significant processing time challenges when dealing with large numbers of features and training examples, leading to increased processing time and potential increases in the upper bound of the training error due to collective calculation of confidence values using the same weights.
Innovation Solution
A modified rule learning method that calculates evaluation values for features, selects features based on these values, calculates confidence values, updates weights, and repeats this process efficiently to generate rule combinations, thereby reducing processing time and maintaining control over the upper bound of the training error.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If AdaBoost collectively calculates confidence values for multiple features using the same weights, then multiple rules can be generated in one iteration, but the upper bound of training error increases and processing time increases
Solution Approach 1:
The patent applies dynamics by updating the weights of training examples sequentially after each confidence value calculation and rule generation. Instead of using static weights for all feature evaluations, the system dynamically adjusts weights based on previous rule outcomes, allowing adaptive reweighting that controls the upper bound of training error while maintaining efficient batch processing
Solution Approach 2:
The patent segments the collective confidence value calculation process into sequential steps where each feature's confidence value is calculated, a rule is generated, and then weights are updated before moving to the next feature. This segmentation allows multiple rules to be generated in one iteration while controlling error accumulation through intermediate weight updates
2Measurement precision
If AdaBoost processes a large number of features and training examples, then more comprehensive rules can be learned, but processing time increases enormously
Solution Approach 1:
The patent maintains continuity of useful action by processing multiple features in sequence within each iteration, continuously generating rules and updating weights without restarting the process. This allows comprehensive processing of large feature sets while maintaining efficient momentum through iterative refinement
Solution Approach 2:
The patent applies preliminary action by calculating evaluation values for all features before selecting which features to process in each iteration. This preliminary evaluation allows the system to prioritize high-value features and process them first, reducing the effective processing burden while maintaining learning accuracy
Data Source
AI summary
A rule learning method for making a computer perform rule learning processing in machine learning includes firstly calculating an evaluation value of respective features in a training example by using data and weights of the training examples; selecting a given number of features in descending order of the evaluation values; secondly calculating a confidence value for one of the given number of selected features; updating the weights of training example, by using the data and weights of the training examples, and the confidence value corresponding to the one feature; firstly repeating the updating for the remaining features of the given number of features; and secondly repeating, for a given number of times, the firstly calculating, the selecting, the secondly calculating, the updating, and the firstly repeating.