XAI-Guided Weighting for Radar ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine-learning algorithms used for classification predictions based on radar measurement datasets face challenges such as being black boxes, making it difficult to explain predictions, and limited accuracy due to small training datasets, which can lead to increased error rates and 'forgetting' during retraining.
Innovation Solution
The method involves using explainable artificial intelligence (XAI) analysis to determine weighting factors for training feature vectors, augmenting training datasets, and employing incremental learning to jointly retrain with initial and additional datasets, ensuring accurate and transparent classification predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine-learning algorithms are used for classification predictions, then prediction capability is improved, but explainability deteriorates (black box problem)
Solution Approach 1:
The patent introduces XAI analysis as an intermediary component between the ML algorithm and the training process. This intermediary analyzes the ML algorithm's predictions and generates weighting factors that guide retraining, thereby maintaining explainability while preserving prediction accuracy. The XAI analysis acts as a mediator that translates the black-box algorithm's behavior into interpretable weighting signals.
2Measurement precision
If training dataset size is increased, then model accuracy is improved, but computational resources and training time increase
Solution Approach 1:
Instead of uniformly treating all training samples, the patent applies local quality by assigning different weighting factors to individual training samples based on XAI analysis. This allows the model to focus computational resources on samples that are most informative for improving accuracy, rather than spending equal time on all samples. The weighting mechanism identifies and prioritizes critical training instances.
Solution Approach 2:
The patent changes the parameter of sample importance by introducing dynamic weighting factors. Rather than treating all training samples equally (uniform parameter), the system adjusts the weight parameter for each sample based on its contribution to model performance, as determined by XAI analysis. This parameter change enables more efficient training by emphasizing high-value samples.
3Measurement precision
If retraining is performed to improve accuracy, then model performance is improved, but forgetting of initial training occurs
Solution Approach 1:
The patent implements feedback by using XAI analysis to evaluate the ML algorithm's current performance and generate weighting factors that guide subsequent retraining. This feedback loop ensures that retraining is directed towards improving weaknesses identified by the XAI analysis while preserving strengths. The system continuously monitors performance and adjusts training weights accordingly, preventing catastrophic forgetting.
Solution Approach 2:
The patent performs preliminary action by analyzing the ML algorithm's behavior before retraining using XAI methods. This preliminary analysis identifies which training samples should be emphasized in retraining, allowing the system to prepare targeted training strategies that improve performance without losing previously learned knowledge. The XAI-guided weighting is established before the actual retraining process begins.
Data Source
AI summary
In accordance with an embodiment, a method of training of a machine-learning algorithm includes: obtaining a training dataset comprising multiple training feature vectors and associated ground-truth labels, the multiple training feature vectors representing respective radar measurement datasets; determining, for each one of the multiple training feature vectors, a respective weighting factor by employing an explainable artificial-intelligence analysis of the machine-learning algorithm in a current training state; and training the machine-learning algorithm based on loss values that are determined based on a difference between respective classification predictions made by the machine-learning algorithm in the current training state for each one of the multiple training feature vectors and the ground-truth labels, wherein the loss values are weighted using the respective weighting factors associated with each training feature vector.