CURE disentangles LLM recommendation circuits into forget-specific, retain-specific, and task-shared modules with tailored update rules to achieve more effective unlearning than weighted baselines.
InInternational Conference on Learning Representations
8 Pith papers cite this work. Polarity classification is still indexing.
8
Pith papers citing it
citation-role summary
baseline 1
citation-polarity summary
years
2026 8roles
baseline 1polarities
baseline 1representative citing papers
Activation steering of SAE-attributed components lets practitioners move from correlational inspection to causal hypothesis testing on CLIP failures, with trust shifting to observed model responses (N=8 experts).
Sparse interventions on 8–64 neurons distributed across layers can activate task behavior in instruction-tuned LLMs, outperforming first-order linear steering approaches by modeling nonlinear neuron interactions.
Dominant error in attribution patching arises from downstream non-linearities; a single HVP correction removes the leading error term and matches Integrated Gradients accuracy at lower cost across 124M-9B models.
A sparse, format-independent Latent Evaluator circuit in mid-to-late MLPs computes judgments across tasks; ablating it removes judgment while leaving world knowledge intact.
Edge-based circuits in vision transformers can be automatically recovered to explain and steer model computations for classification and adversarial behaviors.
ADAG automates description of attribution graphs in language model interpretability by combining gradient-based attribution profiles, a new clustering algorithm, and an LLM explainer-simulator to recover interpretable circuits and identify steerable clusters for jailbreaks.
LRP-based attention head selection and distributed application improve the efficiency and accuracy of function vectors for steering LLMs compared to prior choices.
citing papers explorer
-
CURE:Circuit-Aware Unlearning for LLM-based Recommendation
CURE disentangles LLM recommendation circuits into forget-specific, retain-specific, and task-shared modules with tailored update rules to achieve more effective unlearning than weighted baselines.
-
From Attribution to Action: A Human-Centered Application of Activation Steering
Activation steering of SAE-attributed components lets practitioners move from correlational inspection to causal hypothesis testing on CLIP failures, with trust shifting to observed model responses (N=8 experts).
-
Distributed Sparse Interventions in Language Models
Sparse interventions on 8–64 neurons distributed across layers can activate task behavior in instruction-tuned LLMs, outperforming first-order linear steering approaches by modeling nonlinear neuron interactions.
-
When Attribution Patching Lies: Diagnosis and a Second-Order Correction
Dominant error in attribution patching arises from downstream non-linearities; a single HVP correction removes the leading error term and matches Integrated Gradients accuracy at lower cost across 124M-9B models.
-
Judge Circuits
A sparse, format-independent Latent Evaluator circuit in mid-to-late MLPs computes judgments across tasks; ablating it removes judgment while leaving world knowledge intact.
-
Seeing Through Circuits: Faithful Mechanistic Interpretability for Vision Transformers
Edge-based circuits in vision transformers can be automatically recovered to explain and steer model computations for classification and adversarial behaviors.
-
ADAG: Automatically Describing Attribution Graphs
ADAG automates description of attribution graphs in language model interpretability by combining gradient-based attribution profiles, a new clustering algorithm, and an LLM explainer-simulator to recover interpretable circuits and identify steerable clusters for jailbreaks.
-
Fast & Faithful Function Vectors
LRP-based attention head selection and distributed application improve the efficiency and accuracy of function vectors for steering LLMs compared to prior choices.