Approximate Machine Unlearning through Manifold Representation Forgetting Guided by Self Mode Connectivity

Weiqi Wang, Zhiyi Tian, Chenhan Zhang, Luoyu Chen, Shui Yu

arXiv:2605.22871 · 2026-05-25 공개 · arXiv · PDF

semantic-similarity machine-unlearning representation-space triplet-loss dataset-experiments manifold-representation margin-based-loss model-retraining

Abstract

Machine unlearning is a fundamental mechanism that enforces the right to be forgotten. Existing unlearning studies that rely on label manipulation or task-gradient reversal often deliver limited unlearning effectiveness. Moreover, they can undermine the original learning objective and typically do not guarantee equivalence to standard unlearning by retraining. In this paper, we propose \textbf{ManiF-SMC} (\textbf{Mani}fold \textbf{F}orgetting with \textbf{S}elf \textbf{M}ode \textbf{C}onnectivity), motivated by the observation that a model retrained on the remaining data tends to classify erased samples by their semantic similarity to the retained data. We begin with systematically recasting the approximate unlearning as pushing each erased sample away from its original learned manifold representation centroid toward its nearest semantic neighbors in the retained data. This reformulation aligns unlearning with retraining behavior and operates purely in representation space, reducing reliance on labels and task-specific gradients. To tackle the manifold representation-based unlearning problem, ManiF-SMC encapsulates the unlearning and representation preservation goals in a margin-based triplet loss. Because finding a suitable margin for unlearning is challenging, we propose a self-mode-connectivity module that rapidly reconstructs the local manifold to guide the adaptive margins generation for each unlearning case. Extensive experiments on four representative datasets show that ManiF-SMC achieves unlearning effectiveness comparable to state-of-the-art approximate methods while operating solely within the model's representation space.

한국어 요약

📋 한 줄 요약

**[Approximate Unlearning / Manifold]** ManiF-SMC가 erased sample을 학습된 manifold centroid에서 retained data의 semantic 이웃으로 밀고 self-mode-connectivity로 adaptive margin을 생성 — 라벨·task-gradient 의존 없이 SOTA 수준 unlearning.

🎯 핵심 기여도

💡 핵심 아이디어

Approximate unlearning은 라벨·task-gradient 조작이 아니라 representation space에서의 manifold 작업으로 재정식화될 수 있으며, erased sample을 원 centroid에서 retained semantic 이웃으로 미는 triplet loss + self-mode-connectivity 기반 adaptive margin이 retraining과 정렬된 효과적 unlearning을 만든다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: Unlearning의 라벨 조작 패러다임 탈피, manifold representation 관점으로 일반화 가능한 새 정식화, adaptive margin의 자동화로 실용성. **한계**: 4 데이터셋의 generalization, self-mode-connectivity 모듈의 계산 비용, manifold-based 가정이 일부 task에서 적합도 의존.

🚀 실용적 활용