End-to-End Synthesis of Neural Programs in Weight Space

Published: 05 Mar 2025, Last Modified: 05 Mar 2025ICLR 2025 Workshop Weight Space Learning PosterEveryoneRevisionsBibTeXCC BY 4.0
Track: tiny / short paper (up to 4 pages)
Keywords: weight-space program synthesis, neural subroutines, meta-learning, zero-shot generalization
TL;DR: We present a zero-shot framework that synthesizes a transformer's weights for discrete arithmetic, e.g. (a+b) mod p, achieving near-perfect correctness with no post-hoc fine-tuning.
Submission Number: 43
Loading