LACUNA is a new testbed that injects PII into predefined model parameters to benchmark the localization precision of LLM unlearning methods, revealing that SOTA approaches are imprecise despite strong output performance.
Gradient routing: Masking gradients to localize computation in neural networks
6 Pith papers cite this work. Polarity classification is still indexing.
6
Pith papers citing it
abstract
Neural networks are trained primarily based on their inputs and outputs, without regard for their internal mechanisms. These neglected mechanisms determine properties that are critical for safety, like (i) transparency; (ii) the absence of sensitive information or harmful capabilities; and (iii) reliable generalization of goals beyond the training distribution. To address this shortcoming, we introduce gradient routing, a training method that isolates capabilities to specific subregions of a neural network. Gradient routing applies data-dependent, weighted masks to gradients during backpropagation. These masks are supplied by the user in order to configure which parameters are updated by which data points. We show that gradient routing can be used to (1) learn representations which are partitioned in an interpretable way; (2) enable robust unlearning via ablation of a pre-specified network subregion; and (3) achieve scalable oversight of a reinforcement learner by localizing modules responsible for different behaviors. Throughout, we find that gradient routing localizes capabilities even when applied to a limited, ad-hoc subset of the data. We conclude that the approach holds promise for challenging, real-world applications where quality data are scarce.
citation-role summary
method 1
citation-polarity summary
years
2026 6roles
method 1polarities
use method 1representative citing papers
GRAM selectively trains auxiliary modules so that ablating one at inference removes a targeted capability while preserving the rest, closely tracking data-filtered models at 5x lower cost across 5 capability profiles.
PRISM forms predictions as sparse mixtures of learned prototypes trained with clustering objectives, matching dense model accuracy while enabling ~500x faster data attribution and behavior editing without finetuning.
LIRA aligns latent instruction representations in LLMs to defend against jailbreaks, backdoors, and undesired knowledge, blocking over 99% of PEZ attacks and achieving optimal WMDP forgetting.
A new pipeline uses interpretability to characterize concepts in preference data and shape rewards via feature or data interventions during LM post-training.
ModalImmune enforces modality immunity in multimodal models by controlled collapse of input channels during training using adaptive regularizers and meta-optimization.
citing papers explorer
-
LACUNA: A Testbed for Evaluating Localization Precision for LLM Unlearning
LACUNA is a new testbed that injects PII into predefined model parameters to benchmark the localization precision of LLM unlearning methods, revealing that SOTA approaches are imprecise despite strong output performance.
-
Modular Pretraining Enables Access Control
GRAM selectively trains auxiliary modules so that ablating one at inference removes a targeted capability while preserving the rest, closely tracking data-filtered models at 5x lower cost across 5 capability profiles.
-
Prototype Language Models
PRISM forms predictions as sparse mixtures of learned prototypes trained with clustering objectives, matching dense model accuracy while enabling ~500x faster data attribution and behavior editing without finetuning.
-
Latent Instruction Representation Alignment: defending against jailbreaks, backdoors and undesired knowledge in LLMs
LIRA aligns latent instruction representations in LLMs to defend against jailbreaks, backdoors, and undesired knowledge, blocking over 99% of PEZ attacks and achieving optimal WMDP forgetting.
-
Anatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signal
A new pipeline uses interpretability to characterize concepts in preference data and shape rewards via feature or data interventions during LM post-training.
-
ModalImmune: Immunity Driven Unlearning via Self Destructive Training
ModalImmune enforces modality immunity in multimodal models by controlled collapse of input channels during training using adaptive regularizers and meta-optimization.