Scenario I $\mathrm{for}\mathcal{T}\mathrm{in}\mathcal{D}$ $\underset{\mathrm{for}h\in {\mathcal{H}}_s}{\mathrm{maximize}}$ $\mathsf{P}\left(y\right|x;h)$ subject to ${\mathcal{C}}_s$

→

Scenario II $\mathrm{for}\mathcal{T}\mathrm{in}\mathcal{D}$ $\underset{forh\in {\mathcal{H}}_t}{\mathrm{maximize}}$ $\mathsf{P}\left(y\right|x;h)$ subject to ${\mathcal{C}}_t$