Gains and Losses are Fundamentally Different in Regret Minimization: The Sparse Case | Read Paper on Bytez