CATA enables persistent continual unlearning in VLMs by sign-aware aggregation of unlearning task vectors to suppress conflicts that could revive forgotten knowledge.
An information theoretic approach to machine unlearning
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
years
2026 4verdicts
UNVERDICTED 4roles
background 1polarities
background 1representative citing papers
A new adversarial optimization method for unlearning source-exclusive classes during source-free domain adaptation prevents privacy leakage while preserving target performance.
ICED performs interpretable concept-level unlearning in VLMs by constructing a concept vocabulary via MLLM and decomposing visual representations for targeted optimization.
Jellyfish enables zero-shot federated unlearning through synthetic proxy data generation, channel-restricted knowledge disentanglement, and a composite loss with repair to forget target data while retaining model utility.
citing papers explorer
-
CATA: Continual Machine Unlearning via Conflict-Averse Task Arithmetic
CATA enables persistent continual unlearning in VLMs by sign-aware aggregation of unlearning task vectors to suppress conflicts that could revive forgotten knowledge.
-
$\oslash$ Source Models Leak What They Shouldn't $\nrightarrow$: Unlearning Zero-Shot Transfer in Domain Adaptation Through Adversarial Optimization
A new adversarial optimization method for unlearning source-exclusive classes during source-free domain adaptation prevents privacy leakage while preserving target performance.
-
ICED: Concept-level Machine Unlearning via Interpretable Concept Decomposition
ICED performs interpretable concept-level unlearning in VLMs by constructing a concept vocabulary via MLLM and decomposing visual representations for targeted optimization.
-
Jellyfish: Zero-Shot Federated Unlearning Scheme with Knowledge Disentanglement
Jellyfish enables zero-shot federated unlearning through synthetic proxy data generation, channel-restricted knowledge disentanglement, and a composite loss with repair to forget target data while retaining model utility.