AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

Prototype-based Aleatoric Uncertainty Quantification for Cross-modal Retrieval

Hao Li, Jingkuan Song, Lianli Gao, Xiaosu Zhu, Heng Tao Shen

NeurIPS 2023 · Vision-language and multimodal › Cross-modal learning · approach: Probabilistic embeddings for cross-modal retrieval · 8 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 uncertainty loss (ablation)PAU · MSR-VTT t2vR@1: 45.6 -> 47.0Improved
add uncertainty loss (ablation)PAU · MSR-VTT v2tR@1: 45.2 -> 47.4Improved
add diversity loss on top of uncertainty loss (ablation)PAU · MSR-VTT t2vR@1: 47.0 -> 48.4Improved
add re-ranking with uncertainty vectors to the baseline (ablation)CLIP baseline · MSR-VTT v2tR@1: 45.2 -> 46.0Improved
add re-ranking to full loss model (ablation)PAU · MSR-VTT t2vR@1: 48.4 -> 48.5No clear effect
exponential vs ReLU evidence function (ablation)PAU · MSR-VTT t2vR@1: 46.6 -> 48.5Improved
exponential vs Softplus evidence function (ablation)PAU · MSR-VTT t2vR@1: 47.6 -> 48.5Improved
increase prototypes K from 1 to 8 (ablation)PAU · MSR-VTT t2vR@1: 47.2 -> 48.5Improved
increase prototypes K from 8 to 16 (ablation)PAU · MSR-VTT v2tR@1: 48.3 -> 47.0Hurt
Xavier vs kaiming_uniform prototype initialization (ablation)PAU · MSR-VTT t2vR@1: 47.2 -> 48.5Improved
Xavier vs orthogonal prototype initialization (ablation)PAU · MSR-VTT v2tR@1: 47.1 -> 48.3Improved
train with PAU on small MSVD datasetCLIP baseline · MSVD v2tR@1: 60.1 -> 68.9Improved

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