Domain-Shift Robust Suggestion via Multi-Model Ensemble
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information suggestion systems face challenges in generating robust recommendations across different domains due to domain shift issues, where models trained on one domain do not perform well in another, and existing methods are not effective in selecting the best model without access to data from the target domain.
Innovation Solution
An information processing method that generates a suggested item list by acquiring candidate items from multiple models trained on different domains and selecting items based on prediction values and evaluation values calculated using statistical and probabilistic simulations, ensuring robust performance against domain shifts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a single learning model is trained on data from one domain, then the model can be trained efficiently with available data, but the prediction accuracy decreases when applied to different domains due to domain shift
Solution Approach 1:
The patent divides the learning process into multiple independent learning models, each trained on data from a different domain. Instead of creating one comprehensive model, the system segments the solution into domain-specific models (e.g., model 141 for domain 1, model 142 for domain 2) that can be independently trained and then combined through ensemble methods to achieve both training efficiency and cross-domain reliability.
Solution Approach 2:
The patent changes the parameter of domain specificity by training each learning model on data from different domains rather than using a single domain for all models. This parameter change allows the system to capture domain-specific characteristics while using ensemble techniques to generalize across domains, resolving the contradiction between efficient single-domain training and reliable cross-domain prediction.
2Reliability
If multiple learning models are trained on different domains, then the system becomes robust against domain shift, but the system complexity increases
Solution Approach 1:
The patent merges multiple domain-specific learning models into an ensemble system where the predictions from individual models (141, 142, 143) are combined to produce the final prediction. This merging approach maintains robustness against domain shift while managing complexity through structured combination methods such as averaging or voting mechanisms rather than requiring a single complex model.
Solution Approach 2:
The patent creates a universal ensemble framework that can accommodate multiple domain-specific models with different functions and data sources. The ensemble system serves as a multi-functional container that integrates various learning models trained on different domains, allowing the system to handle diverse data types and domains while maintaining a unified prediction interface that manages overall complexity.
3Object-affected harmful factors
If data from the introduction destination facility is not available, then privacy and security requirements are met, but it becomes difficult to select the best learning model for the target domain
Solution Approach 1:
The patent performs preliminary actions by training multiple learning models in advance on data from various source domains before deployment to the introduction destination facility. These models (141, 142, 143) are pre-trained and stored, allowing the system to make predictions without needing to access or store sensitive data from the target domain, thus maintaining privacy while preserving model selection and adaptation capabilities.
Solution Approach 2:
The patent introduces an intermediary ensemble mechanism that mediates between the need for domain-specific adaptation and data privacy requirements. Instead of directly accessing introduction destination data for model selection, the system uses an intermediary evaluation process that assesses pre-trained models based on their performance characteristics and compatibility with the target domain, enabling model selection without compromising data privacy.
Data Source
AI summary
Provided are an information processing method, an information processing system, and a program capable of generating a suggested item list that is robust against the domain shift by applying a plurality of models that are trained by using datasets of domains different from an introduction destination domain. The information processing system is configured to: acquire one or more candidate items from each of a plurality of models trained by using datasets in one or more domains different from an introduction destination domain; and select, from among a plurality of the acquired candidate items, a plurality of candidate items having different domains from each other as suggested items and generate a suggested item list that is a suggested item list including a plurality of the suggested items and that has robust performance against a domain shift.