Hi folks,
i want to show you some interesting work from Ramin Hasani.
They extrated a generic neural circuit from C. elegans and worked with reinforcement learning to adjust the synapse weight between the neurons.
This is really amazing.
Here are some links:
Short Youtube clip about their work
Nature article
https://www.nature.com/articles/s42256-020-00237-3.epdf
Google Colab Code
Best regardes
Tom
Thanks for this post, very interesting…
Because of this, i discovered probably well-known project, but for me big surprise.
Posting it here, thinking that maybe i am not only one.
Shortly in TED talk
Sorry for late reply.
One of the creators Ramin Hasani is also with the openworm guys working.
I think this could be a good start for implementing explainable AI applications.
For now I didnt had the time to dig into but in the comming month’s I will go trough all their papers.
I will keep you informed!
Open worm is a fascinating project, but I don’t think it does much along the AI line. They’re kind of hoping the AI will emerge from the anatomy, but I don’t think that’s how it works. But if they get there, the sky is the limit.
You can drive a car by using round about 20 neurons wit these system. So this could be a very good way to get it done. But its only a small step.
Pardon? Where does it say that?
In the video…
Probably their nearest shot is to understand brain anatomy starting with small steps. If we aren’t able to really understand 1k or 100k neuron brains how can we aim to understand 100 billion one?
The functionalities of the artificial neurons described in the video seems to be much more advanced than the way neuron cells work in HTM. Just wondering if it is practical to try and deduce brain’s inner working if we don’t use a more accurate cell functionalities?