AI Paper Atlas

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

Compressing Visual-Linguistic Model via Knowledge Distillation

Zhiyuan Fang, Jianfeng Wang, Xiaowei Hu, Lijuan Wang, Yezhou Yang, Zicheng Liu

ICCV 2021 · Label-efficient and robust learning › Knowledge distillation · approach: Few-shot and metric distillation · 85 citations (OpenAlex)

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 hidden embedding (NCE) distillation (ablation)VLP small VL model · COCO CaptioningCIDEr: 110.6 -> 114.4Improved
add attention distribution distillation (ablation)VLP small VL model · COCO CaptioningCIDEr: 110.6 -> 111.8Improved
add both attention and hidden distillation (ablation)VLP small VL model · COCO CaptioningCIDEr: 110.6 -> 115.6Improved
pre-training distillation with all losses (ablation)DistillVLM · COCO CaptioningCIDEr: 115.7 -> 120.1Improved
add fine-tuning classification distillation (ablation)DistillVLM · COCO CaptioningCIDEr: 115.7 -> 117.1Improved
add fine-tuning classification distillation (ablation)DistillVLM · VQA 2.0 test-stdaccuracy: 69.0 -> 69.2No clear effect
add attention or hidden distillation at fine-tuning on top of classification distillation (ablation)DistillVLM · COCO CaptioningCIDEr: 116.3 -> 116.1No clear effect
use last-layer instead of layer-to-layer MSE distillation (ablation)DistillVLM · COCO CaptioningCIDEr: 114.8 -> 115.3Improved
replace MSE with NCE contrastive loss on last layer (ablation)DistillVLM · COCO CaptioningCIDEr: 115.3 -> 115.6Improved
use token-wise embeddings instead of mean-pooled embedding for NCE (ablation)DistillVLM · COCO CaptioningCIDEr: 115.4 -> 115.6Improved
increase NCE queue size from 1 to 4,096 negatives (ablation)DistillVLM · COCO CaptioningCIDEr: 112.5 -> 115.4Improved
align visual tokens using student proposals (teacher not retrained) (ablation)DistillVLM · COCO CaptioningCIDEr: 112.3 -> 115.0Improved

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