Training Neural Networks on Data Sources with Unknown Reliability

When data is generated by multiple sources, conventional training methods update models assuming equal reliability for each source and do not consider their individual data quality during training. However, in many applications, sources have varied levels of reliability that can have negative effects on the performance of a neural network. A key issue is that often the quality of data for individual sources is not known during training. Focusing on supervised learning, this work presents a solution that aims to train neural networks on each data source for a number of steps proportional to the source's estimated relative reliability. This way, we allow training on all sources during the warm-up, and reduce learning on less reliable sources during the final training stages, when it has been shown models overfit to noise. We show through diverse experiments, this can significantly improve model performance when trained on mixtures of reliable and unreliable data sources, and maintain performance when models are trained on reliable sources only.