Improving Conditional Level Generation using Automated Validation in Match-3 Games

https://media.contentapi.ea.com/content/dam/ea/seed/presentations/seed-tog-conditional-level-generation-paper.pdf

EA SEED cVAE conditioned on size, symmetry, and difficulty (median moves of a scripted bot over 30 plays); 51.39% valid levels versus 46.15% in the dataset; difficulty accuracy 33.33%, mean gap 10.19 moves.

Learned level and map generation (PCGRL, MarioGPT, LLM generators)