(a) What is meant by the subdifferential ∂f(x) of a convex function f:Rd→R at x∈Rd ? Write down the subdifferential ∂f(x) of the function f:R→R given by f(x)=γ∣x∣, where γ>0.
Show that x minimises f if and only if 0∈∂f(x).
What does it mean for a function f:Rd→R to be strictly convex? Show that any minimiser of a strictly convex function must be unique.
(b) Suppose we have input-output pairs (x1,y1),…,(xn,yn)∈{−1,1}p×{−1,1} with p⩾2. Consider the objective function
f(β)=n1i=1∑nexp(−yixiTβ)+γ∥β∥1
where β=(β1,…,βp)T and γ>0. Assume that (yi)i=1n=(xi1)i=1n. Fix β2,…,βp and define
κ1=1⩽i⩽n:xi1=yi∑exp(−yiηi) and κ2=i=1∑nexp(−yiηi)
where ηi=∑j=2pxijβj for i=1,…,n. Show that if ∣2κ1−κ2∣⩽γ, then
argminβ1∈Rf(β1,β2,…,βp)=0.
[You may use any results from the course without proof, other than those whose proof is asked for directly.]