Skip to content
AI News HubLIVE
Original source2 min read

Depth-Aware Activation Steering via Attention Residuals

Summary

This paper combines Attention Residuals' learnable cross-layer pathways with Jacobian-lens concept readout into a lightweight depth-aware steering pipeline, showing on Qwen3-0.6B and Qwen3.5-0.8B that distributing a persona vector by depth weights yields stronger internal alignment and behavioral shifts than single-layer injection, alongside findings on geometric decomposition, an intensity window, and metric discipline.

SourcearXiv Computer VisionAuthor: Shilong Zhao (Independent Researcher)
Depth-Aware Activation Steering via Attention Residuals
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions
Read article

[Submitted on 27 Oct 2026]

Title:Depth-Aware Activation Steering via Attention Residuals

View a PDF of the paper titled Depth-Aware Activation Steering via Attention Residuals, by Shilong Zhao (Independent Researcher)

View PDF HTML (experimental)

Abstract:Activation steering is a simple, training-free way to modulate LLM behavior, yet existing methods typically inject vectors at a single layer and lack mechanisms for depth-aware propagation or internal verification. We combine two recent ideas - Attention Residuals (AttnRes) providing learnable cross-layer pathways, and Jacobian-lens (J-space) concept readout - into a lightweight depth-aware steering pipeline. On two small open models (Qwen3-0.6B and Qwen3.5-0.8B), distributing a persona vector according to AttnRes-style depth weights yields stronger internal alignment (deeper-layer concept cosine and rank, p

new | recent | 2026-11

Change to browse by:

cs cs.AI

References & Citations

SciX

Google Scholar

Semantic Scholar

Loading...

Data provided by:

Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer (What is the Explorer?)

Connected Papers Toggle

Connected Papers (What is Connected Papers?)

Litmaps Toggle

Litmaps (What is Litmaps?)

scite.ai Toggle

scite Smart Citations (What are Smart Citations?)

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv (What is alphaXiv?)

Links to Code Toggle

CatalyzeX Code Finder for Papers (What is CatalyzeX?)

DagsHub Toggle

DagsHub (What is DagsHub?)

GotitPub Toggle

Gotit.pub (What is GotitPub?)

Huggingface Toggle

Hugging Face (What is Huggingface?)

ScienceCast Toggle

ScienceCast (What is ScienceCast?)

Demos

Demos

Replicate Toggle

Replicate (What is Replicate?)

Spaces Toggle

Hugging Face Spaces (What is Spaces?)

Spaces Toggle

TXYZ.AI (What is TXYZ.AI?)

Related Papers

Recommenders and Search Tools

Link to Influence Flower

Influence Flower (What are Influence Flowers?)

Core recommender toggle

CORE Recommender (What is CORE?)

Author

Venue

Institution

Topic

About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)

Key points and analysis

Article intelligence

InvestorsAdvanced

Key points

  • Distributing a persona vector across layers using AttnRes-style depth weights outperforms single-layer injection on internal concept cosine and rank (p<0.0001).
  • Effects are sign-specific: reversed vectors invert all six metrics, random directions degrade alignment, a short LoRA fine-tune erases the imprint, and injection rewires attention only downstream (upstream JS = 0.00000).
  • An intensity window exists: effects are monotone over 0.3-0.7, after which generation collapses (617 to 306 to 25 chars).
  • Keyword-rule metrics fail LLM-judge validation (kappa <= 0.09) and two rule-based conclusions were retracted; training depth weights on a steering objective raises final alignment 10x on held-out prompts.

Highlights and analysis are generated automatically and may contain errors. Check the original source.