Based Rejection

Based rejection, a technique where a model selectively refrains from making predictions when confidence is low, is a rapidly developing area of machine learning research. Current efforts focus on improving rejection strategies across diverse applications, from machine translation and radio frequency signal processing to medical diagnosis and financial modeling, employing various architectures including neural networks (e.g., CNNs, MLPs, transformers), support vector machines, and Bayesian methods. This research aims to enhance the reliability and robustness of machine learning systems by mitigating errors stemming from uncertainty or adversarial attacks, ultimately leading to more trustworthy and impactful applications.