Penalty based robust learning with noisy labels