Scientists Create New AI Training Algorithm That Prevents Neural Networks From Breaking Down During Learning
Scientists develop groundbreaking 'manifold Muon' algorithm that prevents neural networks from breaking down during training by constraining weight matrices to maintain unit condition numbers, achieving higher accuracy than existing methods in early tests.