AI Paper Atlas

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

AOEPT: Breaking the Implicit Modality-Reduction Bottleneck in Modality-Missing Prompt Tuning

Jian Lang, Rongpei Hong, Ting Zhong, Fan Zhou

ICML 2026 · Vision-language and multimodal › Cross-modal learning · approach: Multimodal fusion with missing modalities · 0 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 modal-contextualized prompts distilled from training data (vs random prompts) (ablation)AOEPT on CLIP · MM-IMDb, 70% text missingF1-Macro: 48.93 -> 51.50Improved
add instance-aware instantiation of MCPs (ablation)AOEPT on CLIP · MM-IMDb, 70% text missingF1-Macro: 49.17 -> 51.50Improved
add intra-modal latent consistency regularization (ablation)AOEPT on CLIP · MM-IMDb, 70% text missingF1-Macro: 50.56 -> 51.50Improved
replace MCPs with a modality-imputation network trained by reconstruction loss (ablation)AOEPT on CLIP · Food101, 70% text missingAccuracy: 80.77 -> 76.81Hurt
replace MCPs with reconstruction network on HateMemes (ablation)AOEPT on CLIP · HateMemes, 70% text missingAUC: 71.12 -> 70.13Hurt
initialize MAPs prompts with global modality priors from training samples (pilot) (ablation)MAPs · MM-IMDbF1-MacroImproved
use MLP-based instead of attention-based MCP construction (ablation)AOEPT · MM-IMDb, 70% text missingF1-MacroMixed
use initialization-based MCP construction (ablation)AOEPT · MM-IMDb, 70% text missingF1-MacroMixed
increase prompt length and depth up to M=16, N=6 (ablation)AOEPT · MM-IMDb, 70% text missingperformanceMixed
lower the training text-missing rate from 90% to 10% with test rate fixed at 90%MAPs, SyP baselines · MM-IMDbperformanceHurt
lower the training text-missing rate from 90% to 10% with test rate fixed at 90%AOEPT · MM-IMDbperformanceImproved
swap CLIP backbone for single-stream ViLTAOEPT · MM-IMDb, 70% missingF1-Macro avg: 24.45 -> 39.86 over lower boundImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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