Machine-Learned Model for Predicting Saliency Values
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex computer programs, especially black-box models, are difficult to understand due to their complexity, and existing explainable artificial intelligence (XAI) techniques can be computationally expensive.
Innovation Solution
A machine-learned model is trained using input-output-explanation data triples to generate explanations for the inputs as they relate to the output of a computer program, reducing the need for computationally expensive saliency methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If existing XAI techniques are used to explain complex computer programs, then explanations can be generated, but the computational cost and time required increase significantly
Solution Approach 1:
The patent creates a copy of the explanation generation process by training a machine-learned model on input-output-explanation data triples. Instead of running expensive XAI techniques on every new input, the trained model can quickly predict explanations by copying the patterns learned from training data, significantly reducing computational cost while maintaining explanation quality
Solution Approach 2:
The patent performs preliminary action by pre-training the machine-learned model on historical input-output-explanation data before it needs to explain new inputs. This preliminary training allows the model to store explanation patterns in advance, so that when new explanations are needed, the model can quickly retrieve and adapt existing patterns rather than computing explanations from scratch
2Loss of information
If existing XAI techniques are used to explain complex computer programs, then explanations can be generated, but the time required increases significantly
Solution Approach 1:
The trained machine-learned model copies explanation patterns from training data to generate new explanations quickly. This copying approach avoids the time-consuming process of running expensive XAI techniques on every new input, enabling fast explanation generation while maintaining accuracy
Solution Approach 2:
The model performs preliminary learning during training phase by analyzing historical data and storing patterns in its parameters. This preliminary action enables the model to generate explanations rapidly for new inputs by simply applying learned patterns rather than performing time-consuming real-time analysis
3Adaptability or versatility
If complex computer programs are used to provide outputs, then functionality and capability increase, but understandability and interpretability decrease
Solution Approach 1:
The patent introduces an intermediary layer between the complex computer program and the user by training a machine-learned model that translates complex program behavior into simple, interpretable explanations. This intermediary model learns the mapping between inputs, outputs, and explanations, making the complex program's behavior accessible and understandable without changing the program itself
Solution Approach 2:
The trained model creates a simplified copy of the complex program's explanatory behavior. Instead of trying to understand the complex program directly, users can interact with the simpler trained model that copies its explanatory patterns, making the system more interpretable while preserving the original program's functionality
Data Source
AI summary
A computing device accesses a plurality of datasets, each dataset comprising an input of a first plurality of inputs to a computer program, a first output of a plurality of outputs of the computer program, and a value identifying an importance of the input to the first output and produced by an explanation service. The computing device trains a machine-learned model to predict values based on the plurality of dataset. The computing device executes the computer program with a second plurality of inputs to produce a second output. The computing device inserts the second plurality of inputs and the second output into the machine-learned model, and the machine-learned model produces values for the second plurality of inputs based on the second plurality of inputs and the second output.