Learning Value-at-Risk and Expected Shortfall