Estimating and Maximizing Mutual Information for Knowledge Distillation
Estimating and Maximizing Mutual Information for Knowledge Distillation
复制标题
DOI：
10.1109/cvprw59228.2023.00010
复制
发表时间：
2021-10
期刊：
影响因子：
--
通讯作者：
A. Shrivastava;Yanjun Qi;Vicente Ordonez
中科院分区：
文献类型：
--
作者：
A. Shrivastava;Yanjun Qi;Vicente Ordonez
在这项工作中，我们提出了互信息最大化知识蒸馏（MIMKD）。我们的方法使用了一个对比的目标，同时估计和最大限度地提高教师和学生网络之间的局部和全局特征表示的互信息的下限。我们通过大量的实验证明，这可以用来提高低容量模型的性能，从更高性能，但计算昂贵的模型转移知识。这可以用来生成更好的模型，这些模型可以在计算资源较低的设备上运行。我们的方法是灵活的，我们可以提取知识，从教师与任意网络结构的任意学生网络。我们的实证结果表明，MIMKD优于竞争的方法在广泛的学生-教师对不同的能力，不同的架构，当学生网络的容量极低。通过从ResNet-50中提取知识，我们能够在具有ShufflenetV 2的CIFAR 100上从69.8%的基线准确度获得74.55%的准确度。在Imagenet上，我们使用ResNet-34教师网络将ResNet-18网络的准确率从68.88%提高到70.32%（1.44%+）。
In this work, we propose Mutual Information Maximization Knowledge Distillation (MIMKD). Our method uses a contrastive objective to simultaneously estimate and maximize a lower bound on the mutual information of local and global feature representations between a teacher and a student network. We demonstrate through extensive experiments that this can be used to improve the performance of low capacity models by transferring knowledge from more performant but computationally expensive models. This can be used to produce better models that can be run on devices with low computational resources. Our method is flexible, we can distill knowledge from teachers with arbitrary network architectures to arbitrary student networks. Our empirical results show that MIMKD outperforms competing approaches across a wide range of student-teacher pairs with different capacities, with different architectures, and when student networks are with extremely low capacity. We are able to obtain 74.55% accuracy on CIFAR100 with a ShufflenetV2 from a baseline accuracy of 69.8% by distilling knowledge from ResNet-50. On Imagenet we improve a ResNet-18 network from 68.88% to 70.32% accuracy (1.44%+) using a ResNet-34 teacher network.