high-dimensional-projection

Tag

Cards List
#high-dimensional-projection

High-Dimensional Random Projection for Activation Steering in Language Models

arXiv cs.LG · 2026-06-16 Cached

HiDRA is a training-free method that uses high-dimensional random projection for activation steering in LLMs, capturing discriminative signals beyond linear methods and consistently outperforming existing baselines across diverse model families and benchmarks.

0 favorites 0 likes
← Back to home

Submit Feedback