6 comments
Here’s a paper by Floreano at EPFL from 1997 explicitly on Red Queen dynamics for creating neural networks for intelligent robot control.<p>There was lots of discussion of these ideas in the 1990s. In those days we trained very small NNs - tens of nodes - by evolving their weights and topologies. A run could take days on a workstation of the time.<p>This particular paper is about co-evolving predator and prey, where the behavior of each is the ‘evaluation’ of the other.<p><a href="https://infoscience.epfl.ch/entities/publication/a65d0679-6855-4be6-adf4-6c8288092357" rel="nofollow">https://infoscience.epfl.ch/entities/publication/a65d0679-68...</a>
OP's link:
> Now the researchers have addressed this issue by having both the self-improving agent and the evaluator evolve together.<p>and your quote:<p>> This particular paper is about co-evolving predator and prey, where the behavior of each is the ‘evaluation’ of the other.<p>Both sound like the GAN approach that was popularized a decade ago and kinda the start of the "genAI" boom.
> "Instead of improving an agent against a fixed test, we let the evaluation evolve alongside the agent"<p>This quote should have been highlighted earlier in the article.
[dead]
[dead]
Yeah I think it is broadly applicable to tech as a whole, I mean any great startup is just really a counter positioned company to incumbents -