AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

DTO-KD: Dynamic Trade-off Optimization for Effective Knowledge Distillation

Zeeshan Hayder, Ali Cheraghian, Lars Petersson, Mehrtash Harandi, Richard Hartley

ICLR 2026 · Label-efficient and robust learning › Knowledge distillation · approach: Distillation with causal and adaptive guidance

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 changedWhereMeasuredEffect
add dynamic trade-off optimization of task and distillation lossesDeiT-Ti (KD) baseline · ImageNet-1Ktop-1 accuracy (%): 74.5 -> 79.7Improved
use DTO-KD instead of VkDVkD · ImageNet-1K DeiT-Titop-1 accuracy (%): 78.3 -> 79.7Improved
use DTO-KD instead of VkDVkD · ImageNet-1K DeiT-Stop-1 accuracy (%): 82.3 -> 83.1Improved
use 300-epoch DTO-KD instead of 1000-epoch KD baselineDeiT-Ti (KD, 1000 epochs) · ImageNet-1Ktop-1 accuracy (%): 76.6 -> 79.7Improved
use DTO-KD instead of ReviewKD++ReviewKD++ · CIFAR-100 ResNet-56 -> ResNet-20top-1 accuracy (%): 72.05 -> 72.35Improved
use DTO-KD instead of ReviewKD++ReviewKD++ · CIFAR-100 WRN-40-2 -> WRN-40-1top-1 accuracy (%): 75.66 -> 75.68No clear effect
use DTO-KD instead of VkD on ViDT-smallVkD · COCO Swin-smallAP: 48.5 -> 49.6Improved
add dynamic trade-off optimization alone (ablation)no-component KD baseline · COCO ViDT-nanoAP: 41.0 -> 43.1Improved
add projector alone (ablation)no-component KD baseline · COCO ViDT-nanoAP: 41.0 -> 41.8Improved
add projector to DTO (ablation)DTO only · COCO ViDT-nanoAP: 43.1 -> 43.6Improved
add gradient clipping to projector + DTO (ablation)Projector + DTO · COCO ViDT-nanoAP: 43.6 -> 43.7Improved
distill from smaller ViDT-small teacher instead of ViDT-base (ablation)DTO-KD · COCO ViDT-nanoAP: 43.7 -> 43.2Mixed

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy