Particle Network: Thoughts
As I mentioned in the project, ‘Particle Network’ doesn’t even begin to show the potential of what a neural network can do. It is merely a visualization of what happens inside of a neural network and doesn’t solve any problems. It is also missing one of the most important elements of a neural network; the ability to learn and correct itself as it makes mistakes. With that said, I think there are a couple of interesting ideas in Particle Network that are worth exploring.
Visually, I think Particle Network does a good job in representing how our brain functions. According to AI Impacts, the average firing speed of a neuron in our brains is 0.29 seconds. Below is a video that shows Particle Network running much faster than normal .You can begin to see how the nodes continuously fire and interact with each other. Now multiply this by a hundred billion and you will have some idea of how our brain works.
My favorite part of Particle Network is the lack of structure. As far as I understand, and feel free to correct me if I am wrong, traditional neural networks are set up so that the connections between nodes are fixed and don’t change. Outputting particles that gravitate to other nodes depending on their size means that the signal is not being passed between the same two nodes, but rather can go to any node in the system. I can see how this type of neural network can make learning very difficult since there is no path to correct and the entire system is more unpredictable. But perhaps we may be able to channel this unpredictability in a positive way. Forgive me for being naive, but doesn’t creativity derive from the ability to make connections between to unrelated ideas? A neural network where interactions are happening between different nodes each time might be able to make creative leaps more effectively. I would love to solve a creative problem using a traditional neural network and the particle network and compare the results. Perhaps a project for the future…
Leave a Reply