ML Model Evaluation Device for Comprehensive Quality Assessment

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current evaluation techniques for machine learning models require separate assessments for functional and non-functional qualities, making it cumbersome to comprehensively evaluate model quality from various viewpoints, especially during iterative training and re-training processes.

Innovation Solution

An evaluation device that includes an acquirer, a first evaluator for functional quality, a second evaluator for non-functional quality, and a display controller to provide a comprehensive evaluation of machine learning models by acquiring training models, evaluating their robustness, fairness, sufficiency, coverage, and compatibility, and displaying results in a user-friendly format.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple separate evaluations are performed for different quality aspects, then evaluation thoroughness is improved, but evaluation complexity and time consumption increase

Engineering Contradiction:
Improveevaluation thoroughnessVSAvoidevaluation time consumption
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent combines multiple separate evaluation processes (functional quality evaluation, non-functional quality evaluation, training process evaluation, training data selection evaluation, and model selection evaluation) into a single integrated evaluation system. The evaluation device executes all these evaluations simultaneously or in a coordinated manner, allowing comprehensive assessment of model quality from multiple dimensions without requiring separate evaluation sessions, thus reducing total evaluation time while maintaining thoroughness.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The evaluation device is designed as a universal system that can perform multiple types of evaluations through a single interface. It evaluates functional quality, non-functional quality, training process effectiveness, training data quality, and model selection criteria all through one evaluation mechanism, making the evaluation process more efficient and reducing the time required compared to performing each evaluation type separately.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Measurement precision

If comprehensive evaluation from multiple viewpoints is performed, then model quality assessment is improved, but evaluator workload increases

Engineering Contradiction:
Improvemodel quality assessment accuracyVSAvoidevaluator workload
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The patent merges multiple evaluation viewpoints into a unified evaluation framework. The evaluation device simultaneously assesses functional quality, non-functional quality, training process quality, training data quality, and model selection quality, presenting all results through a single interface. This integration reduces the evaluator's workload by eliminating the need to separately conduct and cross-reference multiple evaluation processes while maintaining comprehensive assessment accuracy.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The evaluation device acts as an intermediary that automatically performs and coordinates multiple evaluation tasks. Instead of requiring the evaluator to manually conduct separate evaluations for functional quality, non-functional quality, training processes, and model selection, the evaluation device automates these processes and presents integrated results, significantly reducing evaluator workload while maintaining comprehensive assessment capability.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Adaptability or versatility

If iterative training and re-training processes are performed, then model adaptability is improved, but evaluation and selection complexity increases

Engineering Contradiction:
Improvemodel adaptabilityVSAvoidevaluation and selection complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The evaluation device provides a universal evaluation framework that handles all stages of iterative training and re-training processes. It evaluates functional quality, non-functional quality, training process effectiveness, training data quality, and model selection criteria through a single multi-functional system. This unified approach simplifies the complexity that would otherwise arise from managing multiple separate evaluation processes across iterative training cycles, making the system more manageable while maintaining high model adaptability.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4075301A1Evaluation device, evaluation method, program, and storage medium
Publication Date: 2022.10.19 KK TOSHIBA
  • EP4075301A1 patent drawingFigure 1
  • EP4075301A1 patent drawingFigure 2
  • EP4075301A1 patent drawingFigure 3

AI summary

An evaluation device according to an embodiment includes an acquirer, a first evaluator, a second evaluator, and a display controller. The acquirer acquires a training model that is an evaluation target and evaluation data. The first evaluator evaluates a functional quality of the training model based on output data acquired by inputting the evaluation data to the training model. The second evaluator evaluates a non-functional quality of the training model based on the output data. The display controller outputs an evaluation result screen including a first evaluation result according to the first evaluator and a second evaluation result according to the second evaluator to cause a display device to display the evaluation result screen.