Tracing Learning Sources in Explainable AI Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques are limited in tracing back the learning source responsible for specific behavior in artificial intelligence systems, often only explaining features of input data rather than the actual learning source, making it difficult to correct undesirable behavior.
Innovation Solution
A method and system that generate a learning graph based on similarities among multiple learning sources, allowing iterative comparison of desired and current behaviors across layers of the AI model to detect the responsible learning source, enabling correction of erroneous classifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing explanation techniques are used to trace back classification results, then features of input data can be explained, but the actual learning source cannot be identified
Solution Approach 1:
The patent segments the AI system's learning process into distinct learning sources (e.g., different training datasets, transfer learning sources, adaptive learning environments). By creating separate learning graphs for each source, the system can trace and identify which specific segment contributed to the current behavior, thereby recovering the lost learning source information with high precision.
Solution Approach 2:
The patent introduces learning graphs as intermediary structures that capture and represent the relationship between learning sources and model behaviors. These graphs serve as mediators that bridge the gap between the AI system's internal learning processes and the external observation of model behavior, enabling precise identification of learning sources without direct access to the model's internal memory.
2Adaptability or versatility
If the AI system learns from multiple learning sources, then the system becomes more adaptable and intelligent, but it becomes difficult to isolate and trace the contribution of each learning technique
Solution Approach 1:
The patent divides the multiple learning sources into separate, identifiable segments within the learning graph structure. Each learning source (training data, transfer learning, adaptive learning) is represented as a distinct node or subgraph, allowing the system to maintain high adaptability from multiple sources while enabling easy isolation and measurement of each source's contribution through graph traversal and comparison.
Solution Approach 2:
The patent transforms the complex multi-source learning problem into a different dimension by representing learning sources and their contributions as a graphical structure with multiple layers and nodes. This dimensional transformation from flat data to hierarchical graph enables intuitive detection and measurement of each learning source's impact by analyzing graph topology, node connections, and path similarities.
3Reliability
If the AI system is trained with erroneous data, then the system may exhibit unexpected behavior, but tracing back the specific learning source responsible is very difficult
Solution Approach 1:
The patent implements a feedback mechanism where the system compares the current learning graph against desired behavior specifications. When unexpected or erroneous behavior is detected, the feedback loop triggers a retracing process that uses graph similarity metrics to identify which specific learning source contributed to the error, enabling targeted correction and improving overall system reliability.
Solution Approach 2:
The patent performs preliminary construction of learning graphs during the training process, capturing the contribution of each learning source before erroneous behavior manifests. This preliminary action creates a traceable record of learning provenance that can be quickly queried and analyzed when errors occur, significantly reducing the difficulty of detecting and measuring error sources compared to post-hoc analysis.
Data Source
AI summary
This disclosure relates generally to artificial intelligence system, and more particularly to method and system for tracing a learning source of an explainable artificial intelligence (AI) model. In one example, the method may include receiving a desired behavior of the explainable AI model with respect to input data, generating a learning graph based on similarities among a plurality of learning sources with respect to the input data for the desired behavior and for a current behavior, retracing a learning of the explainable AI model by iteratively comparing the learning graph for the desired behavior and for the current behavior at each of a plurality of layers of the explainable AI model starting from an outer layer, and detecting the learning source responsible for the current behavior based on the retracing.