As an alternative to Deep Learning, some research threads seek to replicate brain network connectivity. I want to explore some modifications to one such project.
What I Want to Accomplish
Add some modules to an existing brain connection implementation in Python using matricies to represent connections. Instead of only using feedforward connections as is done now, I want to “tune” the network using feedback, dampened by inhibition, to isolate resonances associated with different sub-networks in the connections.
In terms of sub-networks, the output should be similar to networks identified from MRI data using recursive modularity analysis. Say, like those from this paper:
Where I Need the Most Help
I need help mostly on the Python code to implement my changes.
Skills with matrix manipulation would be useful, as most connectivity changes would be implemented using matricies.
The project I want to modify is here:
We need to add another one or two stages to the processing. We will modify the existing feedforward architecture to introduce some feedback and inhibition and generate a kind of tuned circuit.
Experience with recursive modularity analysis, or any network analysis really, might be useful.
Also anyone interested in testing the results. Likely that means running it against some cognitively relevant time series prediction benchmarks, like predicting words in a text.
Actual recursive network analysis might be useful to produce graphical representations of structure broken down into hierarchies of the networks found. Something like the tree structures of this earlier vector-based project for structuring natural language:
Why I’m Passionate About This Project
Better understanding AI. 'Nuff said. Very hot research topic at the moment. Fundamental solutions will likely be used in everything: every search engine, every website, every smartphone…
Also nice if you like to understand yourself. What meaning is etc. What happens when you “think”. How freewill is compatible with determinism. What a basis for consciousness might be.