A single$-$layer neural network $($no hidden layer$)$ using the identity function as an activation function will result in linear functional form identical to multiple linear regression, but may generate slightly different predictions than the corresponding multiple linear regression model because the variable weights are determined different methods?