#computational-efficiency

Fast Entropic Approximations cut entropy computation by 37x

Horenko et al. propose non-singular rational approximations of Shannon entropy and KL divergence that preserve mathematical properties while reducing computation cost and improving ML model training.