Research map / Label-efficient and robust learning / Knowledge distillation
Differentiable JPEG-based Input Perturbation for Knowledge Distillation Amplification via Conditional Mutual Information Maximization
SIYU CHEN, Kaixiang Zheng, Ahmed H. Salamah, EN-HUI YANG
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace CE teacher with DJIP teacher (JPEG input perturbation) | KD students · CIFAR-100, ViT-S -> DeiT-T | top-1 accuracy: 69.86 -> 73.97 | Improved |
| replace CE teacher with DJIP teacher | KD, WRN-40-2 -> WRN-16-2 · CIFAR-100 | top-1 accuracy: 74.92 -> 75.64 | Improved |
| replace CE teacher with DJIP teacher | KD, ResNet-110 -> ResNet-20 · CIFAR-100 | top-1 accuracy: 70.67 -> 71.65 | Improved |
| replace CE teacher with DJIP teacher | KD, ResNet-34 -> ResNet-18 · ImageNet | top-1 accuracy: 70.66 -> 71.65 | Improved |
| replace CE teacher with DJIP teacher | AT, ResNet-34 -> ResNet-18 · ImageNet | top-1 accuracy: 70.70 -> 70.78 | No clear effect |
| larger teacher-student capacity gap (different architectures) | DJIP teacher · CIFAR-100 | accuracy gain: up to 2.44% (different) vs smaller gains (same architecture) | Improved |
| add the differentiable JPEG layer to MCMI (fixed centroids) (ablation) | JMCMI · CIFAR-100 WRN-40-2 -> WRN-16-2 KD | top-1 accuracy: 74.92 -> 75.53 | Improved |
| add alternating centroid update on top of JMCMI (ablation) | DJIP · CIFAR-100 WRN-40-2 -> WRN-16-2 KD | top-1 accuracy: 75.53 -> 75.64 | Improved |
| add DJIP JPEG layer to MCMI teacher | DJIP-MCMI · CIFAR-100 VGG-13 -> VGG-8 | top-1 accuracy: 73.83 -> 74.26; gains of up to 0.92% | Improved |
| increase lambda from 0.4 to 1.4 (ablation) | DJIP teacher · ImageNet ResNet-34 -> ResNet-18 KD | student accuracy: 71.362 -> 71.654 | Improved |
| increase lambda from 1.4 to 4.0 (ablation) | DJIP teacher · ImageNet ResNet-34 -> ResNet-18 KD | student accuracy: 71.654 -> 70.694 | Hurt |
| increase quantization rounds from 1 to 5 (ablation) | DJIP teacher · ImageNet ResNet-34 -> ResNet-18 KD | student accuracy: 71.278 -> 71.654 | Improved |
Closest papers by meaning
- Fuse Before Transfer: Knowledge Fusion for Heterogeneous Distillation (ICCV 2025 · similarity 0.87)
- All You Need in Knowledge Distillation Is a Tailored Coordinate System (AAAI 2025 · similarity 0.87)
- Enhancing Logits Distillation with Plug&Play Kendall's $\tau$ Ranking Loss (ICML 2025 · similarity 0.87)
- Adaptive Dual Guidance Knowledge Distillation (AAAI 2025 · similarity 0.87)
- CIFD: Controlled Information Flow to Enhance Knowledge Distillation (NeurIPS 2024 · similarity 0.86)
- Gap Preserving Distillation by Building Bidirectional Mappings with A Dynamic Teacher (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
