p_loss = ALPHA*p_inst + (1-ALPHA)*p_loss        (10)