On Empirical Risk Minimization with Dependent and Heavy-Tailed Data

It allows for the interaction to be even polynomially heavy-tailed, which covers a significantly large class of heavy-tailed models beyond what is analyzed in the learning theory literature.