Can anyone tell me that the range of computational complexity of different ICA algorithms (e.g. FastICA or Infomax or SOBI) in terms of O(K and N) where K is the no. of observations (variables) and N is the no. of samples.
Any reference/link or any suggestion is highly appreciated!
Thank you very much!
The following research paper might be of help to you
http://link.springer.com/chapter/10.1007%2F978-3-642-21326-7_35
I still haven't found any obvious and explicit answer to this query from literature
Table 1 in this paper gives a nice summary along with references. O(d(d+1)n m) where d is the number of variables/dimensions, n is the number of samples, and m is the number of iterations. I think m is the wildcard that has made it hard to find an answer. Convergence can only be proved in special situations. In practice it seems like a fixed m works well, but maybe as the dimension grows m will have to grow too. This will depend on the details of the problem.
http://www.uv.es/~gcamps/papers/Laparra11.pdf
Thanks a lot Greg. This is really helpful and specific answer I was looking for. Cheers
- Badges
- Science topic
- Similar topics
- Biological Science
- Anatomy
- Tissue