In the era of big data, deep learning (DL) techniques have achieved great success across various domains. Notably, label quality is crucial for ensuring both the predictive accuracy and generalization ability of DL models. In many real-world applications, the obtained raw data are normally unlabeled and require annotation by human experts or automated tools. Due to the unexpected errors in automated labeling tools and mis-operation by human annotators, the obtained training data may exist data with incorrect labels. Such label noise would corrupt the training process and mislead the mapping between the feature space and the target space, thereby affecting the performance and reliability of the trained DL model. In this talk, we focus on learning with noisy labels (LNL) in DL and discuss a role-differentiated learning with noisy label (RD-LNL) approach for industrial outlier detection. The proposed outlier detection approach is exploited in a real-world industrial outlier detection task with application to wire arc additive manufacturing.
| 
