Model Learning With Compatibility Loss for Updated Predictions

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for updating machine learning models fail to maintain compatibility between the original and updated models, leading to difficulties in maintaining consistent performance and accuracy.

Innovation Solution

A model learning apparatus and method that extracts specific characteristics from both the original and updated models, using a loss function to align the characteristics and ensure compatibility through machine learning, focusing on prediction errors and other model-specific attributes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a machine learning model is updated using new learning data to improve performance, then the model's adaptability to new environments is improved, but the compatibility with the original model deteriorates

Engineering Contradiction:

Improvemodel adaptabilityVSAvoidmodel compatibility

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent implements feedback by calculating a compatibility index between the updated model and the original model, then using this index to generate feedback signals that guide further model updates. This ensures the updated model maintains compatibility with the original while adapting to new data, resolving the contradiction between adaptability improvement and compatibility preservation

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent changes the loss function parameters to include both accuracy loss and compatibility loss components. By adjusting the weighting of these parameters, the model optimization process balances between improving performance on new data and maintaining compatibility with the original model, thus resolving the contradiction

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If model update is performed using only prediction accuracy evaluation, then the model performance improvement is improved, but the comprehensive compatibility assessment deteriorates

Engineering Contradiction:

Improveprediction accuracyVSAvoidcompatibility evaluation complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the compatibility evaluation into multiple independent components: structure compatibility, parameter compatibility, and performance compatibility. Each component is evaluated separately and then aggregated into an overall compatibility index, making the complex evaluation process more manageable and comprehensive

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20250299101A1Model learning apparatus, model learning method, and program

Publication Date: 2025.09.25 NEC CORP

AI summary

A model learning apparatus of the present disclosure includes: an extracting unit that extracts preset characteristics different from a model prediction error characteristic from a first model generated by machine learning and a second model generated by updating the first model by machine learning; and a learning unit that performs machine learning on the second model by using a loss based on an error between the extracted characteristic of the first model and the extracted characteristic of the second model. Consequently, it is possible to use prediction by a machine learning model for decision making.