Steering by Influence: Curvature Aware Data Weighting for Activation Steering

arXiv:2610.06383 · cs.LG, cs.AI, cs.CL · Submitted 2026-10-05 · Read on arXiv

cs.LG, cs.AI, cs.CL

Submitted: 2026-10-05

Updated: 2026-10-05

Comments: Code: https://github.com/JDIXON-2/Concept_Activation_Transport

Code: https://github.com/JDIXON-2/Concept_Activation_Transport

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Steering — 2× in this paper · explained in 2 episodes
• Activation Steering — 1× in this paper · explained in 4 episodes

Sources

• Concept Heterogeneity-aware Representation Steering
• Influence Functions in Deep Learning Are Fragile
• Discovering Latent Knowledge in Language Models Without Supervision
• The rising costs of training frontier AI models
• Toy Models of Superposition
• GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers
• The Llama 3 Herd of Models
• Studying Large Language Model Generalization with Influence Functions
• LoRA: Low-Rank Adaptation of Large Language Models
• On Effects of Steering Latent Representation for Large Language Model Unlearning
• Improving Activation Steering in Language Models with Mean-Centring
• Scaling Laws for Neural Language Models
• Delta-Influence: Unlearning Poisons via Influence Functions
• RoBERTa: A Robustly Optimized BERT Pretraining Approach
• ASDL: A Unified Interface for Gradient Preconditioning in PyTorch
• The Linear Representation Hypothesis and the Geometry of Large Language Models
• Infusion: Shaping Model Behavior by Editing Training Data via Influence Functions
• Representation Surgery: Theory and Practice of Affine Steering
• Gemma 2: Improving Open Language Models at a Practical Size

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks