SCU: An Efficient Machine Unlearning Scheme for Deep Learning Enabled Semantic Communications

TL;DR

SCU tackles data erasure in DL-enabled semantic communications by enabling joint unlearning of the encoder and decoder while preserving semantic utility. It minimizes the mutual information between the learned semantic representation and erased data $I(Z;D_e)$ and employs a contrastive compensation step that treats erased samples as negatives and the remaining data as positives during retraining. The approach is supported by theoretical analysis and experiments on MNIST, CIFAR10, and CIFAR100, showing improved unlearning effectiveness and efficiency over baselines. This work advances privacy-preserving semantic communication by providing a practical, end-to-end unlearning framework for unsupervised or self-supervised semantic models.

Abstract

Deep learning (DL) enabled semantic communications leverage DL to train encoders and decoders (codecs) to extract and recover semantic information. However, most semantic training datasets contain personal private information. Such concerns call for enormous requirements for specified data erasure from semantic codecs when previous users hope to move their data from the semantic system. {Existing machine unlearning solutions remove data contribution from trained models, yet usually in supervised sole model scenarios. These methods are infeasible in semantic communications that often need to jointly train unsupervised encoders and decoders.} In this paper, we investigate the unlearning problem in DL-enabled semantic communications and propose a semantic communication unlearning (SCU) scheme to tackle the problem. {SCU includes two key components. Firstly,} we customize the joint unlearning method for semantic codecs, including the encoder and decoder, by minimizing mutual information between the learned semantic representation and the erased samples. {Secondly,} to compensate for semantic model utility degradation caused by unlearning, we propose a contrastive compensation method, which considers the erased data as the negative samples and the remaining data as the positive samples to retrain the unlearned semantic models contrastively. Theoretical analysis and extensive experimental results on three representative datasets demonstrate the effectiveness and efficiency of our proposed methods.

Figures (3)

• Figure 2: The overall procedure of SCU. First, we remove the influence of erased data from both semantic encoder $f_\theta$ and decoder $g_\theta$ via minimizing the mutual information between the semantic extracted representation $z_e$ and the specified sample ($i_e$); we name the process as joint unlearning, as shown in red arrows, where $\mathbf{z}_e'$ and $\mathbf{z}_r'$ mean the representations with simulated channel noise. Second, we retrain the unlearned models based on the remaining dataset using contrastive compensation to achieve semantic consistency, as shown in green arrows.
• Figure 3: Evaluations of the impact of different ${\it EDR}$. Model accuracy and backdoor accuracy of downstream models in SCU and VBU decrease as the ${\it EDR}$ increases. HBU easily causes catastrophic unlearning, reflected in huge model accuracy degradation and the highest decoding MSE. Especially in CIFAR10 and CIFAR100, the decoding MSE of HBU is higher than the vertical axis.
• Figure 6: Ablation study when SCU without (abbreviated as w/o) the CC loss on MNIST. It performs similarly to VBU, much worse than the entire SCU.