Liquid Neural Nets and Zero-K

I've been having a look at some interesting videos and papers made by Ramin Hasani.

The subject concerns liquid time-constant neural networks, which are based on reverse-engineering how neurons work in C. elegans.

They are apparently incredibly effective at processing tasks which are both dynamic and occur over a varying time-series. You can see in the video how well it processes info related to autonomous driving compared to our current neural networks.

I'm thinking that Zero-K is a perfect candidate for this sort of neural network, with its dynamic, shifting gameplay. What are your thoughts on this coincidence? If we choose to embrace this technology, how could we apply this to make Zero-K a better experience?

Paper:
https://arxiv.org/pdf/2006.04439.pdf

Video:

Seems interesting. I definitely think the staff should look into this. I imagine it'd allow for the AI to actually compose armies based on what you have.

RTS and neural networks have been getting a lot of attention for a couple years since Starcraft opened up an API for making bots. Instead of a general approach, you might find it interesting to see how people have solved the problem for this space specifically. Here's a paper and their notebook that I think may be more directly applicable.

https://wandb.ai/vwxyzjn/gym-microrts-paper/reports/Gym-RTS-Toward-Affordable-Deep-Reinforcement-Learning-Research-in-Real-Time-Strategy-Games--Vmlldzo2MDIzMTg

https://arxiv.org/pdf/2105.13807.pdf

there "higher up" in zk is really just the person who does the most work. Currently lamer on the zk discord, as the person who made the circuit fork for zk and is the go-to person for any questions and the person "in charge", but if you want to take the initiative and implement this for the zk ai and get something better than circuit you'd quite literally automatically become the glorious supreme prime high lord of utmost power for zk ai