A new study by TSYS School computer scientist Yi Zhou and his colleagues from Jinan University and the University of Exeter points out that recently passed laws and regulations have granted users the right to be "forgotten," which, stated differently, is the right to require data controllers to delete user data. In this regard, various methods for machine unlearning have been proposed to remove individual data points. However, these do not scale to the scenarios where larger groups of features are to be removed. To address this challenge, Zhou et al. propose MAFRO, an optimal-granularity fuzzy decision rule-based classifier that accelerates unlearning via influence functions. Building on granular computing, MAFRO first selects a minimal reduct of attributes, then constructs fuzzy granules with a Gaussian membership function to extract concise decision rules and realizes unlearning through the influence function. "Specifically, instead of training with the full set of attributes, we use the reduct, a minimal subset of attributes that can classify the data with the same accuracy as the full set of attributes. Next, we extract fuzzy rules based on the reduct. Finally, fusing the generated rules establishes the linear model with strongly convex loss functions," Zhou explained. In this way, MAFRO can quantify the divergence caused by attribute deleting and update the model without retraining it, thereby adapting the influence of data removal on the model and accelerating the unlearning process. In the study, which is set to appear in a future issue of IEEE Transactions on Fuzzy Systems, Zhou et al. conduct extensive experiments to evaluate MAFRO on 10 typical datasets in terms of performance and unlearning speed. They then compare MAFRO with the state-of-the-art algorithms. Experimental results demonstrate that MAFRO enhances accuracy by an average of 6.96% and achieves up to 236× speedup for attribute unlearning tasks.
Turner Business note: This publication is Zhou's 15th A-level journal publication since the beginning of 2015.
Comments
Post a Comment