I want to minimize the following with respect to parameters $B$.

$$\sum_{k = 1}^{K} f(A_{k}, B)$$

where $A_k$ are $K$ different data-sets and $B$ is a matrix of parameters.

Can I do this by a gradient descent where:

  1. randomly pick one matrix $A_{i}$ among $K$ matrices.
  2. perform a gradient step that minimizes $f(A_{i}, B)$
  3. repeat steps 1 and 2 until convergence.

The function $f$ is the logarithm of a linear function.

After searching online, this looks like a stochastic gradient descent. Is this really an SGD?

  • $\begingroup$ Yes this is SGD. $\endgroup$
    – Nick Alger
    Jan 15, 2019 at 21:01


Your Answer

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

Browse other questions tagged or ask your own question.