
Zhitao Chen, Zhenxin Cai, Jingyuan Li, Jin Sha · arXiv (Cornell University) 2026 · 2026
DOI: 10.48550/arxiv.2611.15168
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Existing activation steering methods for multimodal large language models (MLLMs) use the global activation mean as the steering vector, which tends to cause cross-task interference and unstable performance. We trace this failure to the multi-centered geometry of task activations. The global mean lands in the empty region between clusters and is faithful to no single task. Moreover, the steering signal in each task-pure local cluster is low-rank, and different query tasks benefit from emphasizing different families of local-basis directions under the shared routing protocol. To address this, we propose MAST (Medoid-Anchored Subspace Steering), a training-free, input-adaptive activation steering framework. MAST groups samples by task, performs multi-center clustering within each group, and selects one representative medoid per cluster. At inference time, each query is routed to multiple representative medoids. For each matched cluster, a low-rank local steering direction is extracted and aggregated across clusters to produce the final steering vector. Building on MAST, MAST-PR (Probe-Reweighted) uses a lightweight task-aware probe to reweight task-specialized local-direction families, yielding a further average improvement of +8.09 points over MAST.
arXiv (Cornell University) 2026 · 0 citations
arXiv (Cornell University) 2026 · 0 citations
Proceedings of the National Academy of Sciences 2026 · 0 citations
arXiv (Cornell University) 2026 · 0 citations
No comments yet — start the discussion below.