Explanatory Program for Machine Learning Reliability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating explanatory information using linear models for machine learning inference results, particularly with graph data, face challenges in reliability due to complex data structures and variations, leading to degraded accuracy and lower reliability of explanatory information.
Innovation Solution
An information processing apparatus and method that generates explanatory information by calculating a ratio of output results for neighborhood data, using the LIME algorithm, and constructing a linear model based on this ratio to output reliable explanations, with specific criteria for neighborhood data generation to ensure high explanatory accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If neighborhood data is generated for explaining machine learning inference results using LIME, then explanatory information can be obtained, but reliability is degraded when data structures are complex (such as graph data) due to difficult control of distribution state
Solution Approach 1:
The patent changes the parameter of neighborhood data generation by controlling the ratio of positive samples (data where class changes) to a specific range (60-80%). This parameter control ensures that the linear approximation model captures the local decision boundary accurately, improving reliability of explanatory information even for complex data structures like graphs.
Solution Approach 2:
The patent performs preliminary action by pre-calculating and controlling the distribution state of neighborhood data before generating explanatory information. By ensuring the ratio of positive samples is within 60-80%, the system prepares optimal conditions for linear model construction, preventing reliability degradation before it occurs.
2Measurement precision
If linear model is constructed using neighborhood data with uncontrolled distribution, then explanatory information can be generated quickly, but accuracy is degraded
Solution Approach 1:
The patent changes the parameter of neighborhood data distribution by controlling the positive sample ratio to 60-80%. This optimized parameter setting improves explanatory accuracy by ensuring the linear model adequately captures local variations, while the automated control mechanism prevents excessive time loss through efficient sampling.
Solution Approach 2:
The patent maintains continuity of useful action by implementing an iterative process that generates neighborhood data, evaluates the positive sample ratio, and adjusts generation parameters continuously. This ensures explanatory accuracy is maintained throughout the process without requiring complete regeneration of data, reducing overall time loss.
3Manufacturing precision
If neighborhood data is generated without controlling the ratio of positive samples, then data generation is simpler, but the linear model accuracy is degraded
Solution Approach 1:
The patent changes the parameter of neighborhood data generation by implementing control over the positive sample ratio (60-80%). This parameter control significantly improves linear model accuracy by ensuring adequate representation of class transitions, while the automated calculation and adjustment mechanisms maintain ease of data generation through programmatic control.
Data Source
AI summary
A recording medium storing an explanatory program for causing a computer to execute an explanatory process. The process includes: generating a plurality of pieces of data based on first data; calculating a ratio of output results, among a plurality of results output in a case that each of the plurality of pieces of data is input to a machine learning model, different from first results output in a case that the first data is input to the machine learning model; generating a linear model based on the plurality of pieces of data and the plurality of results in a case that the calculated ratio satisfies a criterion; and outputting explanatory information with respect to the first results based on the linear model.