arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2107.06993cs.LGcs.CV
收藏

置信条件下的知识蒸馏

Confidence Conditioned Knowledge Distillation

Indian Institute of Science(印度科学学院)

Sourav Mishra, Suresh Sundaram

摘要:

本文提出了一种新的置信度条件知识蒸馏(CCKD)方案,用于将教师模型中的知识转移到学生模型中。现有的最先进的方法采用固定的损失函数用于此目的,并忽略了不同的样本需要转移的不同级别的信息。除此之外,这些方法在数据使用方面也是低效的。CCKD通过利用教师模型分配给正确类的置信度来设计样本特定的损失函数(CCKD-L公式)和目标(CCKD-T公式)来解决这些问题。此外,CCKD通过采用自我调节来阻止这些样本参与学生模型学习更快的蒸馏过程,从而提高了数据效率。对几个基准数据集的实证评估表明,CCKD方法至少达到了与其他最先进方法一样多的泛化性能水平,同时在过程中具有数据效率。通过CCKD方法训练的学生模型不会保留教师模型在训练集上犯下的大部分错误分类。与传统KD方法相比,通过CCKD方法进行的蒸馏提高了学生模型对对抗攻击的弹性。实验表明,MNIST和Fashion MNIST数据集的对抗攻击性能至少提高了3%,CIFAR 10数据集的性能至少提高了6%。

英文摘要:

In this paper, a novel confidence conditioned knowledge distillation (CCKD) scheme for transferring the knowledge from a teacher model to a student model is proposed. Existing state-of-the-art methods employ fixed loss functions for this purpose and ignore the different levels of information that need to be transferred for different samples. In addition to that, these methods are also inefficient in terms of data usage. CCKD addresses these issues by leveraging the confidence assigned by the teacher model to the correct class to devise sample-specific loss functions (CCKD-L formulation) and targets (CCKD-T formulation). Further, CCKD improves the data efficiency by employing self-regulation to stop those samples from participating in the distillation process on which the student model learns faster. Empirical evaluations on several benchmark datasets show that CCKD methods achieve at least as much generalization performance levels as other state-of-the-art methods while being data efficient in the process. Student models trained through CCKD methods do not retain most of the misclassifications commited by the teacher model on the training set. Distillation through CCKD methods improves the resilience of the student models against adversarial attacks compared to the conventional KD method. Experiments show at least 3% increase in performance against adversarial attacks for the MNIST and the Fashion MNIST datasets, and at least 6% increase for the CIFAR10 dataset.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑