AI RESEARCH
Towards Code-Oriented LM Embeddings for Surrogate-Assisted Neural Architecture Search
arXiv CS.LG
•
ArXi:2605.15649v1 Announce Type: new Developing effective surrogates (performance predictors) for Neural Architecture Search (NAS) typically requires expensive fine-tuning or the engineering of complex representations. We propose a low-cost embedding strategy that leverages the inductive bias of Language Models (LMs) to eliminate these overheads. By representing architectures as PyTorch class definition text, we nstrate that off-the-shelf LMs act as competitive feature extractors without NAS-specialized fine-tuning.