Community
cancel
Showing results for 
Search instead for 
Did you mean: 
mo_bri
New Contributor I
1,388 Views

101 Curie Saving Knowledge in NN - Do we need to save and restore cat 0 - committed node 0 and which contexts ?

Jump to solution

Hi,

when saving the knowledge in the 101 hardware NN do I need to save and restore cat 0 ?

Which global contexts do I need to save and restore ?

If the NN supports 127 contexts, then to save all the knowledge in the network is is safest to save all contexts ? This would mean 127 * 128 nodes ( which have 128 bytes for vector data, and 8 byes (4X 16 uint) for cat,contex,inf and min inf = 136 bytes per neuron per context ?

Is there a difference between global context and the context, ie is 127 a special global context ?

Is a context a way to re-use the whole network ? So in essense I want to learn 127 different sets of data,ie images, sounds, patterns etc. So I can use a context for each pattern type, and I am therefore getting 127 neural networks without having to reset and load the knowledge ? What I am trying to ask is how does context work ? What is it ?

Thanks

Marcus

0 Kudos
1 Solution
mo_bri
New Contributor I
55 Views

Hi General Vision were kind enough to answer a few of my questions, so thank Pete for the advice on contacting them, I did and it worked !

So I will add a few answers to each of my threads. Firstly...when we are saving and restoring the NN data (the knowledge), do we need to perform a save on the cat 0 node ?

The answer is there is no node that is for category 0. Cat 0 is used to "untrain" data. If a neuron is firing and giving a false positive (so it should not be firing), then you can take that data that caused the mis-firing and learn it or train ait and rather than pass in a category fpr it, pass in cat as 0. This will exclude this feature vector from the neurons field of interest.

So do not save cat 0 - you can't anyway, as you can only save nodes, not cats.

View solution in original post

6 Replies
idata
Community Manager
55 Views

Hello Marcus,

 

 

Thanks for reaching out!

 

 

First of all I would like to ask you which library you are using. I believe you are using the CurieNeurons but there are two versions of this library, the free version and the pro version. Please keep in mind that if you are using the free version some features will not be available to you.

 

 

That being said, I would like you to provide us some context regarding your project. What are you doing? Are you able to share any additional information?

 

 

Also, I would like you point out some documents that I think will be of help for you:

 

 

https://www.general-vision.com/documentation/TM_CurieNeuronsLibrary.pdf

 

https://www.general-vision.com/documentation/TM_NeuroMem_API.pdf

 

http://www.general-vision.com/documentation/TM_NeuroMem_Technology_Reference_Guide.pdf

 

 

As you may know the CurieNeuron and CurieNeuronsPro libraries were developed by General Vision, so, if you have doubts related to their API, you can contact them ( http://www.generalvision.com/contact-us/). They might be able to clear your doubts on that regard.

 

 

-Peter.
mo_bri
New Contributor I
55 Views

Hi,

I have the pro version from general vision of the lib ,and I also have your free (inter) version of the lib. So I can save and restore the network as the API supports it, I know the free version from General Vision does not support this feature of knowledge persistency.

However, did you read my question? As it states when saving do I need to save committed node 0. So I am already saving the network, suggesting my lib might not support it doesnt even make sense.

Also in terms of giving you more info on my example, why would any further info on what i am doing help you ?I am asking a specific question about how the curie NN is implemented, its got nothing to do with my situation at all, if I told you I am classifying NMist OCR data how would that help for this question ?

Pointing me to the docs is fine, but it doesnt solve my question, so its just adding noise to the thread. If you dont know the answer to my question can you leave it in case someone else can solve it and this thread remains concise and helpful.

Thanks.

mo_bri
New Contributor I
55 Views

Peter,

Haha, You have directed me to a company called General Vision that provides Eye care !! Not the company that made the Curie Hardware NN !

http://www.generalvision.com/contact-us/ http://www.generalvision.com/contact-us/

For those who dont need an eye test (oh the irony), and are interested in who made the module ...

https://www.general-vision.com/ https://www.general-vision.com/

Thanks.

Marcus

idata
Community Manager
55 Views

Hi Marcus,

 

 

I apologize, this was my mistake, I intended to share this link: https://www.general-vision.com/about/contact-us/.

 

 

In this case, since your doubts are related directly to General Vision's module, I suggest you to contact them directly as they might be able to provide you a more accurate answer.

 

 

-Peter.
mo_bri
New Contributor I
56 Views

Hi General Vision were kind enough to answer a few of my questions, so thank Pete for the advice on contacting them, I did and it worked !

So I will add a few answers to each of my threads. Firstly...when we are saving and restoring the NN data (the knowledge), do we need to perform a save on the cat 0 node ?

The answer is there is no node that is for category 0. Cat 0 is used to "untrain" data. If a neuron is firing and giving a false positive (so it should not be firing), then you can take that data that caused the mis-firing and learn it or train ait and rather than pass in a category fpr it, pass in cat as 0. This will exclude this feature vector from the neurons field of interest.

So do not save cat 0 - you can't anyway, as you can only save nodes, not cats.

View solution in original post

idata
Community Manager
55 Views

I'm glad to hear that General Vision was able to provide you useful information, we appreciate that you shared their answers on this community, they will be of much help for other users that intend to use the same software.

 

 

-Peter.
Reply