AI Paper Atlas

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

Progressive Homeostatic and Plastic Prompt Tuning for Audio-Visual Multi-Task Incremental Learning

Jiong Yin, Liang Li, Jiehua Zhang, Yuhan Gao, Chenggang Yan, Xichun Sheng

ICCV 2025 · Vision-language and multimodal › Cross-modal learning · approach: Self-supervised audio-visual representation learning · 1 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 three-stage progressive prompts (TMA, TMDG, TMI) to the prompt-based incremental learnerPHP · AVE/AVVP/AVQA incrementalmean forgetting (F_mean avg): 4.59 -> 3.32Improved
use PHP instead of best baseline PC for transferPHP · AVE/AVVP/AVQA multi-task transfernormalized transfer (Diff): -3.94% -> +7.79%Improved
add TMA adapter alone to baseline (ablation)PHP · AVE/AVVP/AVQA multi-task transfernormalized transfer (Diff): -6.10% -> +8.05%Improved
add TMDG and TMI without the TMA adapter (ablation)PHP · AVE/AVVP/AVQA multi-task transfernormalized transfer (Diff): +7.77% -> -0.71%Hurt
add TMA and TMDG together (no TMI) (ablation)PHP · AVE/AVVP/AVQA incremental (anti-forgetting)mean forgetting (F_mean avg): 4.20 -> 3.21Improved
add all three components versus baseline on mean accuracy (ablation)PHP · AVE/AVVP/AVQA incremental (anti-forgetting)mean accuracy (A_mean avg): 58.20 -> 58.85No clear effect
place components in shallow-middle-deep order (S-M-D) instead of other orderings (ablation)PHP · AVE/AVVP/AVQA incremental (component order)mean accuracy (A_mean): 49.12 -> 58.85Improved
increase deep prompt length (ablation)PHP · AVE->AVQA->AVVP sequenceforgetting vs transferMixed
increase number of task-specific layers (ablation)PHP · AVE->AVQA->AVVP sequenceAVE and AVVP performanceHurt
train tasks incrementally (LLP->AVE) instead of single-task AVE (ablation)PHP · AVEaccuracy: 70.03% -> 73.23%Improved
PHP versus Dualprompt on AVE final accuracyPHP · AVE (incremental)final accuracy: 58.10 -> 55.00Hurt

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