EvoTreeNAD: Genealogy-Guided Evolution for LLM-Driven Neural Architecture Discovery
EvoTreeNAD is introduced, a genealogy-guided evolutionary algorithm that constructs trainable architectures without a supplied seed or a hand-specified search space that outperforms direct generation, best-of-$N$ greedy continuation, and full-family-mean routing.
Li-Shan Yu, D. Jiu, Qi-Zhen Lan et al.
· 0 citations