STEERLSN: Training-Free Representation Alignment through Language-Specific Neuron Subspace Interpolation

Findings of AACL-IJCNLP 2026

Minjun Kim, Dongyeon Kim, Hoyun Song, Jihyo Kim, KyungTae Lim

This paper has been accepted to Findings of AACL-IJCNLP 2026. The paper link, abstract, and results will be added here once the proceedings (ACL Anthology) or a preprint are publicly available.
Keywords Language-specific neurons (LSN) Representation alignment Training-free Neuron subspace interpolation Activation steering Multilingual LLMs
More projects from KAIST MLP Lab →