Entity · technique

Cross Activation Shift Distance

techniqueactivecross-activation-shift-distance-50aad589·1 events·first seen Jun 3, 2026

Aliases: Cross Activation Shift Distance

Co-occurring entities

Backdoor Unlearning Generalization: A Path Toward the Removal of Unknown Triggers in LLMs

More like this (12)

Domain-Shift label-shift assumption Contrastive Activation Addition Actionable Activation Directions for Detecting and Mitigating Emergent Misalignment Across Language Model Families Local Distance Difference Test Cross-Domain Transfer Post-Training Shifts Confidence: A Three-Stage Analysis of How SFT, RL, and OPD Shape Pre-, Intra-, and Post-CoT Calibration IntraShuffler cross-attention adapter Delta Weight Sync Feature-wise Linear Modulation Toward Calibrated Mixture-of-Experts Under Distribution Shift

Recent events (1)

6arXiv · cs.CL·Jun 3, 2026·source ↗

Backdoor unlearning in LLMs generalizes across unknown triggers via cross-backdoor transfer

Researchers demonstrate that training an LLM to unlearn a single backdoor trigger can suppress other backdoors that were never explicitly targeted, a phenomenon they call cross-backdoor transfer. The study spans three model families with backdoors injected via pretraining or continual pretraining, and introduces a new metric called Cross Activation Shift Distance to quantify the relationship between different unlearning interventions. The finding opens a potential defensive strategy where defenders deliberately inject and then remove controlled backdoors to suppress unknown attacker-planted backdoors.

AI Safety Research Alignment and RLHF Cross Activation Shift Distance Backdoor Unlearning Generalization: A Path Toward the Removal of Unknown Triggers in LLMs