Video Compression

VideoNerd

In signal coding there are two famous theorems – Shannon Theorems, which put limits on coding efficiency in lossless and in lossy modes.


However, in Neural Networks there is Cybenko Theorem, which puts constraints on  efficiency of Back-Propagation Neural Network as an approximation of a target function (notice that Neural Network is targeted to approximate functions).

In my words Cybenko Theorem states:
“Each continuous function defined over a compact domain can be approximated with any given precision by sigmoidial functions (or a set of sigmodial functions is dense within a set of continuous functions over a compact domain).”

Consequently if a given function is continuous then with sufficient large hidden layer the function can be approximated with any pre-defined precision.
However, if a given function has at least one discontinuity then back-propagation neural network may totally fails in approximation.

Notice that similar constraint is observed in Fourier Analysis: by means of the trigonometric functions (sin and cos) we can approximate any continuous function defined over a closed interval [a,b]. But if a function has a “jumpy” discontinuity then Gibbs effect appears around the discontinuity point.

 

15 Responses

  1. I’d should examine with you here. Which is not one thing I often do! I get pleasure from reading a post that will make folks think. Additionally, thanks for permitting me to comment!

  2. Good – I should certainly pronounce, impressed with your web site. I had no trouble navigating through all tabs as well as related information ended up being truly simple to do to access. I recently found what I hoped for before you know it at all. Reasonably unusual. Is likely to appreciate it for those who add forums or something, web site theme . a tones way for your customer to communicate. Nice task.

  3. Needed to send you this bit of word to finally thank you very much over again for the remarkable strategies you’ve shown here. This is so tremendously generous with people like you to deliver easily what some people could have made available as an ebook in making some dough for their own end, specifically now that you might well have tried it in case you decided. The strategies in addition acted to become a good way to know that many people have a similar fervor much like my very own to understand a whole lot more with regard to this issue. Certainly there are millions of more pleasurable opportunities in the future for folks who read carefully your blog.

  4. Attractive section of content. I simply stumbled upon your web site and in accession capital to assert that I acquire in fact loved account your blog posts. Any way I will be subscribing to your augment or even I fulfillment you get admission to constantly fast.

Leave a Reply

Your email address will not be published. Required fields are marked *