ALCL: An Adaptive Log-Correntropy Loss for Robust Learning under Non-Gaussian Noise

The article introduces the Adaptive Log-Correntropy Loss (ALCL), a novel loss function designed for robust deep learning in the presence of heavy-tailed and impulsive noise. ALCL adapts its robustness geometry during optimization by jointly learning shape and scale parameters alongside network weights, offering a maximum likelihood formulation with a bounded influence function. Experimental results show that ALCL outperforms traditional mean squared error and static generalized correntropy losses, achieving up to 4.75% improvement in median accuracy under high-noise conditions on benchmark datasets, highlighting its effectiveness for practitioners dealing with non-Gaussian noise in AI applications.