Nonlinear Dynamics

We have acquired more nvidia GPUs inspite of the shortage including:

RTX 3080ti
RTX 3060ti
RTX A4000
RTX A2000

We are ready to resume research infused with this new hardware and hopeful grant extension.

NVIDIA: World Leader in Artificial Intelligence Computing We power the world’s fastest supercomputers.

With our transition to nvidia’s ampere platform, we hope to connect the various regions of our AI together into a larger and more complex model; not unlike the different regions of the brain.

With this work, the AI will be able to interact verbally and visually with objects it “sees” from its “eyes” (video camera ), as well as recognize people who are “close” to “him”. when finished, he should be at the level of a toddler.

At this point, if you’ve read this far, I welcome suggestions on what to name our AI!?

Reply back to the post with names!

We have an alpha stage model successfully running in real time; which takes your portrait into it’s mind via live color video feed (you are encouraged to look around, turn your head, etc to give the AI more data) and then processes it much like we do and then near simultaneously, in a second video frame it outputs what and how it has *learned* to see your live video portrait in it’s surrounding world. To briefly explain, when you first “look” at our AI, a very blurry picture in the right frame appears… but it “learns” fairly quickly and the blurs and splashes of color start to resemble you! It is the differences in your actual portrait to what it “sees and processes” that are the most fascinating!

I intend on uploading a recorded live session in the weeks to follow… stay tuned!

More detailed background for those interested in reading further:

We have a proof of concept model actively running in real time which takes color video as input, and first passes it through a proprietary GAN network which is modeled after the human visual cortex. From there, images or video the AI wants to remember are then stored in another proprietary LSTM (Long Term Short Term [memory]) neural net which at this time is loosely modeled after human short term memory in the neocortex. If the AI “likes” and “approves” what it sees, the video is automatically stored in the AI’s long term memory (also stored in the LSTM).

At this moment, the neural nets are first trained with 400 preprocessed still images of people with varying ages and ethnic backgrounds as well as some wearing glasses and others wearing hats to give the AI a starting “long term memory”. We next perform testing with only 100 images. A starting set of only 500 training and test images is considered very good progress and speed as most commercial, very large “weak” or non-conscious AI’s require literally millions and sometimes more than millions of training images as compared to our 500.

NEXT PHASE aka PHASE II:
Our next goal is very large: to provide absolutely zero training or test images and allow the AI to truly *learn* about first the main people “taking care of it” much like an infant learns who is mommy and who is daddy…. We are very excited about the direction our work is heading and thoroughly enjoy the work we do!

PHASE III:
To model the audio portion of the human brain and start to learn words and sounds and associate those words and sounds with live video sessions. Our hope
Is the AI will learn who “mommy” and “daddy” are and any other people it encounters and then verbally say what/whom it “thinks” it sees. Very exciting research.

If you’ve read this far, and at a minimum are performing graduate level to post-graduate level work/research, I welcome potential volunteers to collaborate with. I have an nVidia grant as well as won various NSF super computer grants. Please just send a message and we can at least see if we are a match. This is a very rare and fascinating opportunity!

High Level Architecture of a GAN