Approximate Machine Unlearning through Manifold Representation Forgetting Guided by Self Mode Connectivity
Researchers have introduced ManiF-SMC, a novel approach to machine unlearning that aims to improve effectiveness and preserve original learning objectives. This method reformulates unlearning as pushing erased data points away from their learned representations towards semantically similar retained data. ManiF-SMC utilizes a triplet loss within the representation space and incorporates a self-mode-connectivity module to adaptively guide the unlearning process. Experiments demonstrate that ManiF-SMC achieves unlearning effectiveness comparable to existing state-of-the-art methods. AI
IMPACT This new unlearning technique could enhance data privacy compliance for AI systems by offering more effective and less disruptive data removal.