Project summary

With the recent explosion in interest of the Drosophelia Connectome I was disappointed to see that a lot of projects online simply swap the neurons with perceptrons and call it a day. Using the fly's brain structure to justify saying they thought a fly to for example...trade stock. This is untrue and it is an overstatement of what is being done. This is standard dense neural network practice. It is merely no different than your regular run of the mill AI...just reshuffled a little. I decided to explore this graph structure of the fly with a biophysical perspective. I want to see what I can contribute to science and I already have some experience in interpretability of DNA. Therefore I will be using a MORGAN style cosine dampened network to see if we can learn anything about the fly. I already started at https://github.com/CK5515/Drosophelia_Connectome

What are this project's goals? How will you achieve them?

My goal is to rigorously arrive at a conclusion where I can answer the question whether we can map this biophysical neural network onto a fly and what can we interpret from the results. This may branch into other fields of AI interpretability or understanding neuroscience. I will achieve my goals by publishing all my experiments and findings to a public repository for everyone to read.

How will this funding be used?

At this moment in current time I am limited to an rtx3060. This limits the scale at which I can work at and makes it slow to iterate. Right now I was only able to probe a subcircuit of the fly brain which limits the conclusions I can arrive at. To scale as something full-time...

Living expenses of around 2000eur monthly. This allows maximum attention on the project.

Cloud compute of around 1000 to 3000 euro is what I aim for. The scale will depend on scoping tests early on so these funds will go only towards gpu instances to transition from subcircuits.

I also want to be fully transparent that I am also applying to MATS Incubation (for the stipend/support) and BlueDot Rapid Grant (infrastructure). If these go through I will scale the Manifund request accordingly.

Who is on your team? What's your track record on similar projects?

This is a solo project. I work on interpretable DNA models at SWAEV Genomics @ https://swaev.com . This fly project is an outlet for me to work on research I can actually publish and not keep as some trade secret. This is essentially my first public facing research and I want to be transparent throughout. I post failures and successes. We are supported by NVIDIA Inception, GCP for Startups and Enterprise Ireland New Frontiers.

What are the most likely causes and outcomes if this project fails?

If this project fails it still answers questions. This specific architecture is not explored and is a gap that if filled will at least show data to researchers that this does not work... or what parts of it did work. Either way it is progress and these architectures are under explored.

How much money have you raised in the last 12 months, and from where?

None, this is a fresh experiment.