Active Learning Instance Selection via Input Space Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Active learning in machine learning models tends to bias training instances towards instances close to the determination boundary, leading to a lack of improvement in model accuracy, affecting not only rule-based models but all machine learning models.
Innovation Solution
An information processing apparatus that divides the instance input space of multiple machine learning models into regions, assigns probabilities to these regions, calculates sampling probabilities based on these probabilities, and selects instances for labeling and retraining, thereby reducing bias and improving model accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If active learning is performed by selecting instances close to the determination boundary, then the model can focus on uncertain predictions, but the training instances become biased against the distribution and model accuracy cannot be improved
Solution Approach 1:
The patent applies local quality by dividing the instance input space into multiple regions with different properties. Each region is assigned a probability value reflecting its importance for model improvement. This allows the system to treat different regions of the input space differently, selecting instances from regions with higher probabilities while maintaining overall distribution balance, thus resolving the contradiction between focusing on uncertain predictions and maintaining training instance distribution stability
2Reliability
If instances are selected intensively from regions close to the determination boundary, then the model can improve predictions in uncertain areas, but the training instances become biased and overall model accuracy improvement is limited
Solution Approach 1:
The patent segments the instance input space into multiple regions based on model prediction characteristics. By calculating probabilities for each region and using these to guide instance selection, the system can improve prediction reliability in uncertain areas while maintaining adaptability to the overall data distribution. The segmentation allows balanced improvement across different regions rather than intensive focus on a single boundary region
Data Source
AI summary
An information processing apparatus of the present disclosure includes: a region dividing unit that divides an instance input space of each of a plurality of machine learning models into a plurality of regions and assigns a probability to each of the division regions; a probability calculating unit that calculates a sampling probability on a predetermined instance belonging to the division region based on the probability assigned to the division region; and an instance selecting unit that selects the predetermined instance based on the sampling probability on the predetermined instance.