From Wulfram Gerstner
Optimizing the return by policy-gradient in a multi-step environment naturally leads to eligibility traces. A few important mathematical steps are sketched here.
