DynamicKD: An Effective Knowledge Distillation via Dynamic Entropy Correction-Based Distillation for Gap Optimizing
This work addresses a specific bottleneck in knowledge distillation for improving model compression, though it appears incremental as it builds on existing entropy-based methods.
The paper tackles the performance gap between teacher and student networks in knowledge distillation by proposing a dynamic entropy correction method that adjusts the student's output entropy in real-time, achieving a 2.64-point accuracy improvement over traditional distillation and 0.87 points over state-of-the-art CRD on CIFAR100.
The knowledge distillation uses a high-performance teacher network to guide the student network. However, the performance gap between the teacher and student networks can affect the student's training. This paper proposes a novel knowledge distillation algorithm based on dynamic entropy correction to reduce the gap by adjusting the student instead of the teacher. Firstly, the effect of changing the output entropy (short for output information entropy) in the student on the distillation loss is analyzed in theory. This paper shows that correcting the output entropy can reduce the gap. Then, a knowledge distillation algorithm based on dynamic entropy correction is created, which can correct the output entropy in real-time with an entropy controller updated dynamically by the distillation loss. The proposed algorithm is validated on the CIFAR100 and ImageNet. The comparison with various state-of-the-art distillation algorithms shows impressive results, especially in the experiment on the CIFAR100 regarding teacher-student pair resnet32x4-resnet8x4. The proposed algorithm raises 2.64 points over the traditional distillation algorithm and 0.87 points over the state-of-the-art algorithm CRD in classification accuracy, demonstrating its effectiveness and efficiency.