AI Paper Atlas

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

Module-wise Adaptive Distillation for Multimodality Foundation Models

Chen Liang, Jiahui Yu, Ming-Hsuan Yang, Matthew Brown, Yin Cui, Tuo Zhao, Boqing Gong, Tianyi Zhou

NeurIPS 2023 · Label-efficient and robust learning › Knowledge distillation · approach: Rethinking distillation objectives · 4 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 OPTIMA module-adaptive distillation to layerwise distillation (CoCa-Tiny12)CoCa-Tiny12 · COCO CaptionCIDEr: 116.8 -> 121.2Improved
Add OPTIMA (CoCa-Tiny12)CoCa-Tiny12 · NLVR2 test-paccuracy: 80.4 -> 81.3Improved
Add OPTIMA (CoCa-Tiny12)CoCa-Tiny12 · VQA test-stdaccuracy: 71.2 -> 71.6Improved
Add OPTIMA (CoCa-Tiny12)CoCa-Tiny12 · SNLI-VE testaccuracy: 83.9 -> 84.3Improved
Add OPTIMA (CoCa-Tiny6)CoCa-Tiny6 · COCO CaptionCIDEr: 112.2 -> 113.5Improved
Add OPTIMA (CoCa-Tiny6)CoCa-Tiny6 · NLVR2 test-paccuracy: 76.5 -> 77.0Improved
Distill fixed arm Img+Txt+Multi instead of Img+Multi (ablation)CoCa-Tiny · NLVR2accuracy: 81.1 -> 80.4Hurt
Replace fixed-arm and random-arm distillation with OPTIMA arm selection (ablation)CoCa-Tiny · multimodal tasksstudent performanceImproved
Use OPTIMA instead of layerwise distillationCoCa-Tiny · dev setprediction loss and KL discrepancyImproved
Reward based only on KD loss (r_KD) instead of combined reward (ablation)OPTIMA · NLVR2accuracyMixed
Reward based only on layerwise losses (r_LWD) instead of combined reward (ablation)OPTIMA · COCO CaptionCIDErMixed
Vary EMA halflife of reward distribution too large or too small (ablation)OPTIMA · student performanceaccuracyHurt

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