Machine Learning

Chapter 8

Answers to Exercises

Section 1.1

2. Let $ϵ>0$ be a proper threshold, and let smax y be the second maximum of y. Furthermore, assume that $h(y)=0$ indicates the absence of decision (rejection). Then for all $i=1,\dots ,n$ choose $h_i(y)=[\max y-\mathrm{smax}y>ϵ]\cdot [i=\mathrm{a}\mathrm{r}\mathrm{g}\mathrm{m}\mathrm{a}\mathrm{x}y]$. This gets rid of the robustness issue. Clearly, the choice of ϵ, which isn't ...