AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Prophet Attention: Predicting Attention with Future Attention

Fenglin Liu, Xuancheng Ren, Xian Wu, Shen Ge, Wei Fan, Yuexian Zou, Xu Sun

NeurIPS 2020 · Vision-language and multimodal › Human and embodied modeling · approach: Neural image captioning · 25 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 Dynamic Prophet AttentionAoANet · MSCOCOCIDEr: 129.6 -> 133.4Improved
add Dynamic Prophet AttentionGVD · Flickr30k EntitiesCIDEr: 59.9 -> 62.7Improved
add Dynamic Prophet AttentionGVD · Flickr30k EntitiesF1loc: 11.8 -> 15.5Improved
add Dynamic Prophet AttentionUp-Down · MSCOCOCIDEr: 123.5 -> 129.0Improved
use DPA (noun-phrase windows, masked non-visual words) instead of CPA (ablation)Prophet Attention · MSCOCOCIDEr: 126.3 -> 128.1Improved
mask Prophet Attention on non-visual words (ablation)DPA · MSCOCOCIDEr: 0.5 CIDErImproved
increase regularization lambda from 0.1 to 0.3 (ablation)DPA · MSCOCOCIDEr: 129.0 -> 121.3Hurt
increase lambda from 0.1 to 1 (ablation)DPA · MSCOCOCIDEr: 129.0 -> 114.8Hurt
lower lambda from 0.1 to 0.05 (ablation)DPA · MSCOCOCIDEr: 129.0 -> 128.1Hurt
use L1 attention loss instead of L2 (ablation)DPA · MSCOCOCIDEr: 128.2 -> 129.0Improved
use L1 attention loss instead of KL divergence (ablation)DPA · MSCOCOCIDEr: 126.9 -> 129.0Improved
add DPA to paraphrase modelLSTM-Attention · MSCOCO paraphraseBLEU: 29.2 -> 36.5Improved

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