Evolutionary Strategies

What I worked on

Explored the idea of evolutionary strategies inspired by biology. Looked at curiosity and survival as drivers of exploration.

What I noticed

• Evolutionary algorithms rely on variation and selection
• Neuroevolution mutates weights instead of gradients
• Curiosity can emerge without explicit reward
• MAP-Elites and open-ended algorithms encourage diverse solutions
• The free energy principle links to minimizing surprise

”Aha” Moment

That evolution and curiosity-driven systems can learn without explicit goal signals.

What still feels messy

How to connect these biological principles to practical ML training loops.

Next step

Experiment with a small neuroevolution setup and simple energy dynamics.