there is another typo in the paper, in Equation (2). The time indices in the update should be shifted. I think the best way to write it would be
M(t) = (1-eta)*(log(f1(t) + n(t)) + eta * M(t-1)
Then, M(t) should be compared with f2(t). Please check your implementation.
Thanks Idan and Ameer to point this out!