More info for noise adder please

The front end.
Post Reply
charlie
Posts: 60
Joined: Wed Aug 09, 2017 10:55 pm

More info for noise adder please

Post by charlie » Thu Mar 21, 2019 10:54 pm

Could anyone provide more information on noisy-adder such as a paper by Adam Zagorecki and Marek Druzdzel "The noisy-average model for local probability distribution" (which is on Researchgate but I don't have an access without an edu email).

I am quite confused by GeNIe's noisy-adder node's behavior especially with the complication of leak and weight. It doesn't behave the way I assumed.

Many thanks

Charlie

marek [BayesFusion]
Site Admin
Posts: 270
Joined: Tue Dec 11, 2007 4:24 pm

Re: More info for noise adder please

Post by marek [BayesFusion] » Mon Mar 25, 2019 4:43 pm

The Noisy-Adder model is described in the doctoral dissertation of Adam Zagorecki (2010), Section 5.3.1 Non-decomposable Noisy-average. Here is where you should be able to find it:

http://d-scholarship.pitt.edu/6542/

I hope this helps,

Marek

charlie
Posts: 60
Joined: Wed Aug 09, 2017 10:55 pm

Re: More info for noise adder please

Post by charlie » Tue Mar 26, 2019 11:03 pm

Thanks.

charlie
Posts: 60
Joined: Wed Aug 09, 2017 10:55 pm

Re: More info for noise adder please

Post by charlie » Sun Mar 31, 2019 4:27 am

Hi Marek

Forgive my ignorance - I still feel confused with noisy adder's behavior. Please see attached a simple noisy-adder net and a spreadsheet version with calculations in cells colored in red. The CPTs in the net are different from those in the Excel sheet. Could you please enlighten me where I got wrong?

Many thanks

Charlie
Attachments
NoisyAdder.xlsx
(10.74 KiB) Downloaded 35 times
NoisyAdder.xdsl
(1.39 KiB) Downloaded 30 times

charlie
Posts: 60
Joined: Wed Aug 09, 2017 10:55 pm

Re: More info for noise adder please

Post by charlie » Mon Apr 01, 2019 12:18 am

I know now where I got wrong - I didn't exclude the distinguished states when calculating the average.

Recalculated with the exclusion is attached (tab "Simple_correct"), which is the same as network's CPT.

While the problem is solved I'm still wondering why distinguished states should be excluded. Any enlightening on this is still appreciated, please.

Charlie
Attachments
NoisyAdder_1.xlsx
(13.12 KiB) Downloaded 31 times

Post Reply