where \(X\) is a closed convex set, may be reduced to the experts problem, as long as \(X\) is equipped with an optimisation oracle for linear functions \cite{kale} \cite{kun}. We map one expert to each linear inequality \(a_i x\ \leq b_i\). Given a probability assignment \(p_i\) for each expert \(i\), the no-regret algorithm's mixed strategy for the current turn, we solve