Alexandre Lemire Paquin

New loss function symmetrization improves neural network robustness to label noise

Researchers have developed a new method for training neural networks that is more robust to errors in labeled data. This approach, called symmetrization of loss functions, theoretically guarantees better performance whe…