.png){kind=icon}
.png){kind=icon}
Return
Optimization of Adversarial Reprogramming for Transfer Learning on Closed Box Models
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2025.3552219.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In this work, we optimise a transfer learning approach for predicting the Remaining Useful Life (RUL) of ball bearings, particularly in scenarios with limited data availability. Accurate RUL prediction is crucial for improving maintenance strategies, reducing downtime and improving machine reliability, making it highly relevant to industry. We use the Black Box Adversarial Reprogramming (BAR) algorithm to process target domain data in a source domain model through adversarial reprogramming. While it has been shown that this concept can work well in image classification, and a further modification has been developed to classify time series features, this work focuses primarily on the remaining key challenges such as the selection and comparison of appropriate loss functions, the optimisation of hyperparameters using Bayesian methods, and data labelling in the absence of ground truth. Our results show an increase in the performance of the BAR algorithm on the macro f1 score of 0.23 on the training set and up to 0.21 on the test set.
Keywords:
Bars
Time series analysis
Transfer learning
Prediction algorithms
Classification algorithms
Closed box
Vectors
Training
Feature extraction
Predictive maintenance
Adversial reprogramming
closed box transfer learning
machine learning
RUL-classification