3 ms·
I'm fascinated that brain is compartmentalized enough, despite all cells being connected to each other, so a "local circuitry" can be identified for a specific
by sedatk 4y ago
I'm fascinated that brain is compartmentalized enough, despite all cells being connected to each other, so a "local circuitry" can be identified for a specific action. I'd presume all cells would be firing more or less for any action. The existence of circuitry like structures makes it less organic and less mysterious for some reason. :)
Now I wonder if artificial neural networks exhibit the same circuitry-like structures.
- feet 4y agoNot all cells connect to each other remember and there are many different cell types with different functions which form functions specific circuits. Cell functionality goes far beyond the simple excitation-inhibition model
- sedatk 4y agoNot directly, but they all must at least be connected indirectly, right?
- feet 4y agoHow many degrees of separation are there between cells? Many of them are pretty distant from each other even if they are technically part of the same overall network, we also have to consider the directionality of neuron signalling
- red75prime 4y agoThey do. For example, training transformers on addition creates circuits that perform modular addition. See "A Mechanistic Interpretability Analysis of Grokking"
- lowbloodsugar 4y agoHaving learned about computers before learning about the brain I assumed that the “grey matter” in our head was a general purpose substrate. However the brain is very much a set of heterogeneous processors with well defined purposes. We have evolved for a limited number of purposes, sadly.