Adaptive Machine Learning Explainability With Knowledge Graph Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional explainability techniques for machine learning models are limited in explaining model behavior over time, failing to account for changes due to data and system factors, such as new entities or relationships, leading to inconsistent predictions.
Innovation Solution
Adaptive explainability is achieved through a system that leverages a knowledge graph, variational autoencoder, and optimal transport plan to provide feedback on changing relationships and entities, allowing for context-aware explanations of model predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional explainability techniques are used to explain machine learning model predictions, then basic model behavior can be explained, but the explanations fail to account for changes over time due to new entities and relationships in the knowledge base
Solution Approach 1:
The patent implements dynamic explainability by continuously updating the knowledge graph and re-running explainability algorithms as new entities and relationships are added. The system adapts to temporal changes by maintaining an evolving knowledge base that reflects current domain understanding, allowing explanations to remain accurate over time despite changing data landscapes
Solution Approach 2:
The system incorporates feedback loops where model predictions are continuously evaluated against the knowledge graph, and discrepancies trigger updates to the knowledge base. This feedback mechanism ensures that the explainability system learns from new patterns and entities, improving both reliability and temporal adaptability of explanations
2Loss of information
If detailed domain knowledge is incorporated into the explainability system to improve explanation quality, then context-sensitive explanations are achieved, but the system complexity increases due to knowledge graph maintenance and processing
Solution Approach 1:
The patent segments the knowledge graph into modular domains and relationships, allowing independent processing and maintenance of different knowledge subsets. This segmentation reduces overall system complexity by enabling targeted updates and parallel processing, while still maintaining comprehensive context information across all domains
Solution Approach 2:
The system introduces an intermediary layer that maps complex knowledge graph structures to simplified explanation templates. This intermediary transforms detailed domain knowledge into human-readable explanations without requiring the entire knowledge graph to be processed directly, reducing computational complexity while preserving context information
3Measurement precision
If the knowledge graph is continuously updated to reflect new entities and relationships, then explanation relevance is improved, but the processing time and computational resources increase
Solution Approach 1:
The patent implements periodic batch updates of the knowledge graph combined with incremental updates for critical changes. Rather than continuous full re-processing, the system performs explainability analysis at scheduled intervals and only re-processes affected portions when specific triggers occur, maintaining high relevance while reducing overall processing time
Solution Approach 2:
The system pre-processes and indexes knowledge graph data in advance, organizing it into query-optimized structures before explainability analysis is needed. This preliminary action allows rapid retrieval and processing during actual explanation generation, improving relevance without proportionally increasing processing time
Data Source
AI summary
One or more computing devices, systems, and/or methods for providing adaptive explainability for machine learning models are provided. A knowledge structure, representing entities with nodes and relationships between entities as edges between the nodes, is processed to create knowledge system entity embeddings. A dimensionality of the knowledge system entity embeddings is reduced to create dimensional embeddings. The dimensional embeddings and relationships are processed using an optimal transport plan to generate feedback. The feedback is used to modify the knowledge structure for generating adaptive explainability information that explains predictions generated by the machine learning models.