AI Paper Atlas

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

Beyond Instance-Level Alignment: Dual-Level Optimal Transport for Audio-Text Retrieval

Wenqi Guo, Shikui Tu, Lei Xu

ICLR 2026 · Vision-language and multimodal › Cross-modal learning · approach: Probabilistic embeddings for cross-modal retrieval

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 reliability-aware marginals (RAM) (ablation)DART (ResNet38+BERT) · AudioCapsaudio-to-text R@1: 54.44 -> 55.27Improved
add reliability-aware marginals (RAM) (ablation)DART (BEATs+BERT) · AudioCapsaudio-to-text R@1: 71.1 -> 72.1Improved
add reliability-aware marginals (RAM) (ablation)DART (ResNet38+BERT) · Clothoaudio-to-text R@1: 22.48 -> 23.54Improved
add reliability-aware marginals (RAM) (ablation)DART (ResNet38+BERT) · Clothotext-to-audio R@1: 17.30 -> 17.18No clear effect
use correlation alone as marginal instead of full RAM (ablation)DART · AudioCaps (batch 64)mean R@1: 44.92 -> 44.35Hurt
use full RAM (corr, variance, kurtosis) instead of uniform marginals (ablation)DART · AudioCaps (batch 64)mean R@1: 44.92 -> 45.55Improved
use kurtosis-only marginals instead of uniform (ablation)DART · AudioCaps (batch 64)mean R@1: 44.92 -> 45.29Improved
add feature-level loss L_UWD to L_IOT (ablation)DART · AudioCaps (batch 128)audio-to-text R@1: 52.45 -> 53.08Improved
use L_UWD alone without instance-level loss (ablation)DART · AudioCaps (batch 128)audio-to-text R@1: 53.08 -> 0.10Hurt
replace matching/IOT loss with DART feature-level extensionIOT loss · ESC-50 zero-shotR@1: 79.25 -> 80.75Improved
replace Luong et al. with DART under 40% unlabeled dataLuong et al. 2024 · AudioCaps semi-supervised 40% unlabeledtext-to-audio R@1: 28.58 -> 33.24Improved
replace Luong et al. with DART under 40% noisy captionsLuong et al. 2024 · AudioCaps noisy labels 40%audio-to-text R@1: 34.37 -> 37.09Improved

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