This is the equation given to me in the lectures: $$S=\frac1N\sum_{n=1}^N(\mathbf{x}_n-\overline{\mathbf{x}}) (\mathbf{x}_n-\overline{\mathbf{x}})^T,$$

which doesn't make sense to me when I think about it. The $\mathbf{x}_n$ are $D$ dimensional vectors for $D$ features. So subtracting the mean will again result in a vector. Then taking the transpose multiplication will result in a scalar. So as far as I can see, this equation will end up as a scalar instead of a matrix.

How would I be able to correctly interpret this equation? I know there are different equations in matrix form, but I should use this equation.

  • 10
    $\begingroup$ These are column vectors, not row vectors. $\endgroup$
    – whuber
    Commented Oct 14, 2021 at 18:56
  • 4
    $\begingroup$ What you have is not a covariance equation, it is a matrix, (cf. whuber's comment) of random variables, and the expectation of the matrix (which is just the matrix of expectations of the matrix entries), is the covariance matrix. $\endgroup$ Commented Oct 14, 2021 at 19:21

1 Answer 1


Let's write this in matrix forms:

$$\overline{\mathbf{x}} = \left[\matrix{\bar{x}_1\\\bar{x}_2\\\vdots\\\bar{x}_d}\right] = \frac{1}{N}\sum_{i=1}^N\mathbf{x}_i = \frac{1}{N}\sum_{i=1}^N\left[\matrix{x_{i,1}\\x_{i,2}\\\vdots\\x_{i,d}}\right] $$

So the (biased) estimate of the covariance matrix is a square matrix, like below:

$$S=\frac1N\sum_{i=1}^N(\mathbf{x}_i-\overline{\mathbf{x}}) (\mathbf{x}_i-\overline{\mathbf{x}})^T = \frac1N\sum_{n=1}^N \left(\left[\matrix{x_{i,1}\\x_{i,2}\\\vdots\\x_{i,d}}\right]-\left[\matrix{\bar{x}_1\\\bar{x}_2\\\vdots\\\bar{x}_d}\right]\right) \left(\left[\matrix{x_{i,1}\\x_{i,2}\\\vdots\\x_{i,d}}\right]-\left[\matrix{\bar{x}_1\\\bar{x}_2\\\vdots\\\bar{x}_d}\right]\right)^T =\\ \frac1N\sum_{i=1}^N \left[\matrix{x_{i,1}-\bar{x}_1\\x_{i,2}-\bar{x}_2\\\vdots\\x_{i,d}-\bar{x}_d}\right] \left[\matrix{x_{i,1}-\bar{x}_1&x_{i,2}-\bar{x}_2&\cdots&x_{i,d}-\bar{x}_d}\right]=\\ \frac1N\sum_{i=1}^N \left[\matrix{(x_{i,1}-\bar{x}_1)^2 & (x_{i,1}-\bar{x}_1)(x_{i,2}-\bar{x}_2) & \cdots & (x_{i,1}-\bar{x}_1)(x_{i,d}-\bar{x}_d) \\ (x_{i,1}-\bar{x}_1)(x_{i,2}-\bar{x}_2) & (x_{i,2}-\bar{x}_2)^2 & \cdots & (x_{i,2}-\bar{x}_2)(x_{i,d}-\bar{x}_d) \\ \vdots & \vdots & \ddots & \vdots \\ (x_{i,1}-\bar{x}_1)(x_{i,d}-\bar{x}_d) & (x_{i,2}-\bar{x}_2)(x_{i,d}-\bar{x}_d) & \cdots & (x_{i,d}-\bar{x}_d)^2}\right] $$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.