Mitigating Unintended Memorization in Language Models Via Alternating Teaching

• Published in: ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) pp. 1 - 5
• Main Authors: Liu, Zhe, Zhang, Xuedong, Peng, Fuchun
• Format: Conference Proceeding
• Language: English
• Published: IEEE 04.06.2023
• Subjects: automatic speech recognition; Data models; Data privacy; knowledge distillation; Language modeling; Predictive models; Privacy; Signal processing; Speech recognition; Training; unintended memorization
• DOI: 10.1109/ICASSP49357.2023.10096557

Recent research has shown that language models have a tendency to memorize rare or unique sequences in the training corpora which can thus leak sensitive attributes of user data. We employ a teacher-student framework and propose a novel approach called alternating teaching to mitigate unintended memorization in sequential modeling. In our method, multiple teachers are trained on disjoint training sets whose privacy one wishes to protect, and teachers' predictions supervise the training of a student model in an alternating manner at each time step. Experiments on LibriSpeech datasets show that the proposed method achieves superior privacy-preserving results than other counterparts. In comparison with no prevention for unintended memorization, the accuracy loss is small when training records are sufficient.