Robustness-Reinforced Knowledge Distillation With Correlation Distance and Network Pruning

📅 2023-11-23
🏛️ IEEE Transactions on Knowledge and Data Engineering
📈 Citations: 3
Influential: 0
📄 PDF

career value

210K/year
🤖 AI Summary
Existing knowledge distillation (KD) methods rely heavily on Kullback–Leibler (KL) divergence, which suffers from insufficient or biased knowledge transfer under high- or low-entropy teacher distributions; moreover, standard data augmentation can inadvertently degrade KD performance. To address these issues, we propose a robust KD framework comprising three key innovations: (i) replacing KL divergence with correlation-based distance to mitigate entropy sensitivity; (ii) integrating structured network pruning to enhance student model robustness; and (iii) identifying and mitigating the adverse interference of data augmentation in KD via a multi-stage teacher–student co-optimization strategy. Extensive experiments on CIFAR-100, FGVC-Aircraft, TinyImageNet, and ImageNet demonstrate state-of-the-art performance: student models achieve average accuracy gains of 1.2–2.7% over prior methods and exhibit significantly improved resilience to input noise and adversarial perturbations.
📝 Abstract
The improvement in the performance of efficient and lightweight models (i.e., the student model) is achieved through knowledge distillation (KD), which involves transferring knowledge from more complex models (i.e., the teacher model). However, most existing KD techniques rely on Kullback-Leibler (KL) divergence, which has certain limitations. First, if the teacher distribution has high entropy, the KL divergence's mode-averaging nature hinders the transfer of sufficient target information. Second, when the teacher distribution has low entropy, the KL divergence tends to excessively focus on specific modes, which fails to convey an abundant amount of valuable knowledge to the student. Consequently, when dealing with datasets that contain numerous confounding or challenging samples, student models may struggle to acquire sufficient knowledge, resulting in subpar performance. Furthermore, in previous KD approaches, we observed that data augmentation, a technique aimed at enhancing a model's generalization, can have an adverse impact. Therefore, we propose a Robustness-Reinforced Knowledge Distillation (R2KD) that leverages correlation distance and network pruning. This approach enables KD to effectively incorporate data augmentation for performance improvement. Extensive experiments on various datasets, including CIFAR-100, FGVR, TinyImagenet, and ImageNet, demonstrate our method's superiority over current state-of-the-art methods.
Problem

Research questions and friction points this paper is trying to address.

Improving student model performance via robust knowledge distillation
Addressing KL divergence limitations in knowledge transfer efficiency
Mitigating adverse effects of data augmentation in distillation
Innovation

Methods, ideas, or system contributions that make the work stand out.

Uses correlation distance for knowledge distillation
Incorporates network pruning to enhance efficiency
Integrates data augmentation for improved robustness
🔎 Similar Papers
No similar papers found.
S
Seonghak Kim
School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST)
G
Gyeongdo Ham
School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST)
Y
Yucheol Cho
School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST)
D
Daeshik Kim
School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST)