Multi-Predictor Bandit Algorithm for Option Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing bandit algorithms, such as those described in Non-patent Literature 1, face challenges in maximizing the accumulation of observation values due to deviations in the learned relation between observation values and contexts, leading to suboptimal option selection and increased losses over trials.
Innovation Solution
An information processing apparatus and method that acquires relevant information for multiple options, determines the best option using multiple predictors that learn independently from training data, and accumulates observation values and relevant information to improve selection accuracy, allowing for the use of the most appropriate predictor for each option.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single predictor is used to learn the relation between observation values and contexts, then the device complexity is reduced, but the accuracy of option selection deteriorates when the learned relation deviates from the original relation
Solution Approach 1:
The patent divides the single prediction task into multiple independent predictors, each learning the relation between observation values and contexts from different perspectives or with different assumptions. This segmentation allows the system to handle cases where the original relation deviates by selecting the most appropriate predictor for each situation, thereby improving reliability without significantly increasing overall complexity.
Solution Approach 2:
The patent changes the parameters of the prediction system by introducing multiple predictors with different learning configurations rather than using a single predictor. This allows the system to adapt to varying conditions and relations between observation values and contexts, improving the accuracy of option selection when the original relation model is insufficient.
2Reliability
If multiple predictors are used to independently learn the relation between relevant information and observation values, then the accuracy of option selection is improved, but the device complexity increases
Solution Approach 1:
The patent employs multiple predictors beyond what a single predictor could provide, accepting the increased complexity as a necessary trade-off for improved accuracy. The system uses this excess predictive capacity to ensure reliable option selection even when the learned relation deviates from the original model, applying partial or excessive action to overcome the limitations of a single predictor.
Solution Approach 2:
The patent introduces multiple predictors as intermediaries between the input data (relevant information and contexts) and the final option selection. Each predictor acts as an intermediary that processes the data through different learning pathways, and the system selects the most appropriate intermediary (predictor) for each decision, thereby improving reliability while managing complexity through structured mediation.
3Adaptability or versatility
If the learned relation between observation values and contexts deviates from the original relation, then the adaptability to new patterns is improved, but the consistency with the original model deteriorates, leading to suboptimal option selection
Solution Approach 1:
The patent makes the prediction system dynamic by allowing the selection of different predictors based on the situation. When the learned relation deviates from the original model, the system can dynamically switch to a predictor that has adapted to the new pattern, while maintaining the option to use predictors consistent with the original model when appropriate. This dynamic approach resolves the contradiction between adaptability and consistency.
Solution Approach 2:
The patent creates a composite prediction system where multiple predictors with different characteristics are combined. Some predictors are designed to be consistent with the original model, while others are more adaptable to deviated relations. By combining these diverse predictors into a single system, the patent achieves both adaptability to new patterns and consistency with the original model, avoiding suboptimal option selection.
Data Source
AI summary
In order to further increase accumulation of observation values when selecting an option from a plurality of options for which probability distribution is unknown, an information processing apparatus (1) includes an acquisition unit (11), a determination unit (12), and an accumulation unit (13). The acquisition unit (11) acquires pieces of relevant information respectively associated with a plurality of options. The determination unit (12) determines an option to be selected from among the options. The accumulation unit (13) accumulates an observation value of a gain obtained from the determined option and relevant information of the option in a storage apparatus as training data. The determination unit (12) determines, with use of any of a plurality of predictors, an option to be selected from among the options. The plurality of predictors independently learn a relation between the relevant information and the observation value with reference to the training data.