Other than the book “Rebooting AI” by Gary F. Marcus and Ernest Davis, probably a somewhat orthogonal place to begin is an article by Dr. Robert Epstein here titled, “The Empty Brain”. With these ideas in mind, we set out to develop a mathematical model of a neuron in action in the present moment, not really as a computer or information processor but as something possibly different. The idea was to develop something simple and then observe its behavior in action as an experiment. In fact without any signals entering, the neuron does nothing and has no memory at all. It might have some attributes that have developed over time, or even perhaps appear pre-wired, but it is nothing like a CPU or computer RAM. These are in some ways “empty” neurons to borrow a phrase from Dr. Epstein’s article. But how empty are newly developed neurons prior to experience and the flow of synaptic signals and without training?
(This is a simplified example for demonstration purposes only.)
Pictured here is a snapshot of the J-Neuron mathematical model during a self driving simulation using CARLA. We chose this simulation as a fun example that is easy to understand. In this case, the interest is not in developing self driving car software in its entire form but instead focus on the core math model processing the real time video feeds. In this way we can explore new ideas by using the CARLA simulator to observe how part of an AI system learns at a fundamental level of the J-neuron. Instead of a training ground for machine learning, this is more of an experimental arena where we tap into the J-Neuron math models and observe what is going on “under the hood” in order to find ways to improve those models and develop new approaches. In the image to the left from the camera sensor input there are two search regions shown outlined in cyan. Within these regions the green dots show where J-neurons processing that portion of the image are firing in that instant in time. The car steering signal is shown in the darker blue rectangle extending to the left, however this is not important. In the images on the right is an example of the real-time history of the signals entering, interacting within, and exiting one of the J-Neuron (math model). Notice how the Dendrites act as filters, the Soma acts like a nonlinear amplifier, and the Axon acts like a high pass filter followed by an amplifier. There are features of the model that create a memory like effect in the moment. In each frame of the video, these signals update and it gets quite exciting at times. Probably the most interesting thing is that this J-Neuron acts much more like an analog device than digital. It is so tempting to assign a digital computer like analogy to the brain because the firing of the neurons seems so much like a digital 1 or 0. But when we look inside at the interacting parts of a J-Neuron everything is analog signals interacting in the moment with nothing stored. This is why we liked the “Empty Brain” article by Dr. Epstein so much. But once connected into a system – exactly how empty the brain begins is something we just don’t know. Is the brain prewired in some way (such as a language system) as described in the book, ”The Birth of the Mind“ by Gary Marcus? Do the mathematics of these J-Neurons lend to some sort of preference of structure, and how dynamic or even reconfigurable are connections in such a network? These are the exciting challenges we are researching.
The proprietary mathematical model underlying the J-neurons is still under development in its third year. It is entirely a closed form analytical model which provides easy access to explicit derivative information without numerical approximation. We think the best way to avoid this is to focus on a single neuron and its quantum aspects with no goal of solving a specific problem. Instead, we focus on a hybrid architecture of classical and quantum mathematical modeling of a single neuron, the J-Neuron.
Contact p@joab.io for more information, including early access to the J-Neuron API.
© Copyright 2024 JOAB, All Rights Reserved.