search

Agnuxo1 / QBOX

QBOX: A Three-Dimensional Optical Neural Network for Efficient Information Processing Francisco Angulo de Lafuente
https://www.researchgate.net/profile/Francisco-Angulo-Lafuente-3/interest
MIT License
0 stars 0 forks source link

CUBE_REPORT import missing #1

Open lonnietc opened 3 days ago

lonnietc commented 3 days ago

Hello,

I am trying to run your QBOX and other repos as this is very interesting but there seems to be missing the custom CUBE_REPORT import:

import CUBE_REPORT  # Import the custom report module

I would like to replicate this:

https://www.youtube.com/watch?v=nZy38NFyDOk

Also interested in trying to apply QBOX to LLMs and Spiking neural networks if possible.

Where can I get this? Thanks

Agnuxo1 commented 3 days ago

Good morning. The basis of my work is to create new neural network models that are faster and more efficient. To do this, I have transformed conventional neurons into points of light that work through simulated optical physics with Raytracing. I am also working on a more efficient memory and state saving format for the neural network using a hologram. You can run the QBOX program by adding a blank sheet for the CUBE_REPORT since the report is not essential to the operation of the program. Although I recommend you try more advanced versions of my work as QBOX is one of the first demos of the concept. Below is the list of works chronologically.

https://github.com/Agnuxo1/QBOX https://github.com/Agnuxo1/QuBE https://github.com/Agnuxo1/NEBULA https://github.com/Agnuxo1/NEBULA-EVOLUTION https://github.com/Agnuxo1/Holography_Raytracing

If you have any questions, do not hesitate to contact me. Greetings: Francisco Angulo de Lafuente

El mié, 16 oct 2024 a las 1:25, LonnieTC @.***>) escribió:

Hello,

I am trying to run your QBOX and other repos as this is very interesting but there seems to be missing the custom CUBE_REPORT import:

import CUBE_REPORT # Import the custom report module

Where can I get this? Thanks

— Reply to this email directly, view it on GitHub https://github.com/Agnuxo1/QBOX/issues/1, or unsubscribe https://github.com/notifications/unsubscribe-auth/BHS2SU5EXUI6MPCJVHJG2CLZ3WP6NAVCNFSM6AAAAABQAGQQPKVHI2DSMVQWIX3LMV43ASLTON2WKOZSGU4TAMJUHE3TGNI . You are receiving this because you are subscribed to this thread.Message ID: @.***>

--

lonnietc commented 2 days ago

Thanks for the quick reply.

I have been reading the information that you have on QBOX and other repositories and it is very interesting to me as I think that it could offer great potential for Optical Neural Networks (ONN) and even think that just maybe it could be modified to have spiking neural network properties as well which might further enhance it.

I think that the Optical domain has a great potential for neural networks and have been researching them for some time with my latest interest in "Nanophotonic media for artificial neural inference" (2019) by ERFAN KHORAM, et. al. in which they use glass to develop a neural network that uses virtually no power except for the input image.

I like the idea of a evolutionary spiking ONN based upon QBOX ray-tracing ideas such that the network evolves, maybe like your Nebula-Evolution (Nebula9999) ideas which I would like to see more implementation if possible.

The goal was to start with QBOX (try to replicate the video) to get a good feel for it and then work through QuBE, Nebula, Nebula-Evolution, and into Holography_Raytracing to understand all that you are working on now.

If all goes well then just maybe there might be a potential to make some actual neuromorphic hardware based upon D2NN or the Glass Nanophotonic media approach but that is much further down the road from here.

Thanks again and have a great day

Agnuxo1 commented 2 days ago

Thank you very much for your interest in the project.

Your comments are very insightful. Initially, I envisioned developing a photonic processor, but that remains a distant prospect – almost science fiction. Therefore, I decided to simulate one using ray tracing and a holographic system based on simulated optical physics.

I believe that even with a simple simulation, we can improve the efficiency of neural networks. By representing neurons as simulated points of light, the GPU can process billions of them simultaneously. I have already successfully run models with several billion light-based neurons on my computer.

I am currently working on simplifying the prototype demonstration programs and uploading new repositories. I will be uploading further repositories to the platform tomorrow.

Sincerely,

Alternatively, a slightly less formal closing:

Thank you very much for your interest in the project.

Your comments are very insightful. Initially, I envisioned developing a photonic processor, but that remains a distant prospect – almost science fiction. Therefore, I decided to simulate one using ray tracing and a holographic system based on simulated optical physics.

I believe that even with a simple simulation, we can improve the efficiency of neural networks. By representing neurons as simulated points of light, the GPU can process billions of them simultaneously. I have already successfully run models with several billion light-based neurons on my computer.

I am currently working on simplifying the prototype demonstration programs and uploading new repositories. I will be uploading further repositories to the platform tomorrow.

Best regards,

El mié, 16 oct 2024 a las 21:19, LonnieTC @.***>) escribió:

Thanks for the quick reply.

I have been reading the information that you have on QBOX and other repositories and it is very interesting to me as I think that it could offer great potential for Optical Neural Networks (ONN) and even think that just maybe it could be modified to have spiking neural network properties as well which might further enhance it.

I think that the Optical domain has a great potential for neural networks and have been researching them for some time with my latest interest in "Nanophotonic media for artificial neural inference" (2019) by ERFAN KHORAM, et. al. in which they use glass to develop a neural network that uses virtually no power except for the input image.

I like the idea of a evolutionary spiking ONN based upon QBOX ray-tracing ideas such that the network evolves, maybe like your Nebula-Evolution (Nebula9999) ideas which I would like to see more implementation if possible.

The goal was to start with QBOX (try to replicate the video) to get a good feel for it and then work through QuBE, Nebula, Nebula-Evolution, and into Holography_Raytracing to understand all that you are working on now.

If all goes well then just maybe there might be a potential to make some actual neuromorphic hardware based upon D2NN or the Glass Nanophotonic media approach but that is much further down the road from here.

Thanks again and have a great day

— Reply to this email directly, view it on GitHub https://github.com/Agnuxo1/QBOX/issues/1#issuecomment-2417747391, or unsubscribe https://github.com/notifications/unsubscribe-auth/BHS2SU62MN2MKIOZIQJJZZ3Z3234DAVCNFSM6AAAAABQAGQQPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMJXG42DOMZZGE . You are receiving this because you commented.Message ID: @.***>

--

lonnietc commented 1 day ago

Hello

I have been exploring your repositories and was wondering if you were going to release the code to run tetris, snake, and pong like in the video that I posted earlier?

Also, will you have any small LLM demos coming out since I would like to see how some of that could be done with the QBOX approach since that also has potential to offer highly distributed spiking neural networks. I was playing with the idea of a highly distributed (P2P) LLM that is massive in size but could be more efficient than current implementations which use transformers and RNN's (SpikeGPT and RWKV).

Also if you would like to have off-list discussions then do you have a Discord or Slack channel? If not, then I have a small Slack channel that I can send you an invite if I have your email.

I think that there could be huge potential with the work that you are doing and I would like to discuss it more with you if you are interested.

Have a great day

Agnuxo1 commented 1 day ago

Good morning. Write me directly to my email and we'll talk. Greetings: @.***

El jue, 17 oct 2024 a las 19:25, LonnieTC @.***>) escribió:

Hello

I have been exploring your repositories and was wondering if you were going to release the code to run tetris, snake, and pong like in the video that I posted earlier?

Also, will you have any small LLM demos coming out since I would like to see how some of that could be done with the QBOX approach since that also has potential to offer highly distributed spiking neural networks. I was playing with the idea of a highly distributed (P2P) LLM that is massive in size but could be more efficient than current implementations which use transformers and RNN's (SpikeGPT and RWKV).

Also if you would like to have off-list discussions then do you have a Discord or Slack channel? If not, then I have a small Slack channel that I can send you an invite if I have your email.

I think that there could be huge potential with the work that you are doing and I would like to discuss it more with you if you are interested.

Have a great day

— Reply to this email directly, view it on GitHub https://github.com/Agnuxo1/QBOX/issues/1#issuecomment-2420100786, or unsubscribe https://github.com/notifications/unsubscribe-auth/BHS2SU5ZA374MDRGD2KNDS3Z37XI5AVCNFSM6AAAAABQAGQQPKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMRQGEYDANZYGY . You are receiving this because you commented.Message ID: @.***>

--