search

hunar4321 / find-global-maximum

Simulation of life like characters and several learning algorithms to test the ability of each algorithm to enable the characters reaching global maximum with minimum effort
https://www.brainxyz.com/
MIT License
17 stars 5 forks source link

Questions regarding the simulation #1

Open BradKML opened 1 year ago

BradKML commented 1 year ago
  1. Is assortative mating and genetic heritability considered (for behaviors)?
  2. Can the liberal vs conservative divide be seen as hyperparameter optimization "learning rate"? (see here and here)
  3. Has CSR selection been considered ("r/K selection" as reproductive rate and resource intensity) as an intermediary factor?
  4. Are multi-layered cycles considered? It might be that certain behaviors are anti-fragile against the different waves (K-Wave + Kuznet and Cliodynamics)
hunar4321 commented 1 year ago

Hi

  1. Mating is an interesting concept. Neat algorithm has done it, I guess. For the first part of the video I used a simple hill climbing algorithm (no mating). A bare form version of the algorithm in python can be found here: https://github.com/hunar4321/genetic-algorithm/blob/master/very_simple_GA.py

I also made a blog explaining the algorithm here: https://www.brainxyz.com/machine-learning/genetic-algorthim/

  1. Yes, the steps are similar to the learning rate in SGD
  2. No
  3. I don't know about that, I'll read about it thanks for the links
BradKML commented 1 year ago

Regarding CSR selection, it is based on Grime & Pierce and later Winemiller & Rose on how resource scarcity and competition changes selected traits and genetics. I think some crass "pundit" also made similar speculative theories for human behavior. Can the mechanism of fight for resource vs fleeing be created in these simulation as well? https://en.wikipedia.org/wiki/Lotka%E2%80%93Volterra_equations