Research map / Vision-language and multimodal / Cross-modal learning
Anchor-Guided Gradient Alignment for Incomplete Multimodal Learning
Zhi-Hao Guan, Longfei Huang, Yang Yang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add multimodal instance retriever (MIR) in place of dummy values for missing modality (ablation) | ANGA (ViLT) · HateMemes (70% text-missing) | AUROC: 59.38 -> 64.63 | Improved |
| add MIR retrieval (ablation) | ANGA (ViLT) · Food101 (70% text-missing) | ACC: 68.53 -> 74.86 | Improved |
| add anchor-guided gradient alignment (GA) on top of MIR (ablation) | ANGA (ViLT) · HateMemes (70% text-missing) | AUROC: 64.63 -> 66.82 | Improved |
| add gradient alignment on top of MIR (ablation) | ANGA (ViLT) · MM-IMDb (70% text-missing) | F1-Micro: 54.17 -> 56.32 | Improved |
| add semantic-enhanced adapter (SEA) on top of MIR + GA (ablation) | ANGA (ViLT) · HateMemes (70% text-missing) | AUROC: 66.82 -> 68.54 | Improved |
| add SEA on top of MIR + GA (ablation) | ANGA (ViLT) · Food101 (70% text-missing) | ACC: 77.01 -> 77.23 | Improved |
| replace RAGPT with ANGA | retrieval-based reconstruction · HateMemes (70% text-missing) | AUROC: 64.10 -> 68.54 | Improved |
| replace RAGPT with ANGA | retrieval-based reconstruction · Food101 (70% text-missing) | ACC: 75.53 -> 77.23 | Improved |
| vary number of retrieved instances K from 1 to 9 (ablation) | ANGA · HateMemes and MM-IMDb | AUROC / F1-Micro | No clear effect |
| increase cosine threshold tau of the gradient cone (ablation) | ANGA · HateMemes and MM-IMDb | AUROC / F1-Micro | Mixed |
| train with higher missing ratio (10% to 50%), test at 90% missing (ablation) | ANGA and baselines · HateMemes | AUROC | Improved |
Closest papers by meaning
- Post-pre-training for Modality Alignment in Vision-Language Foundation Models (CVPR 2025 · similarity 0.88)
- Amend to Alignment: Decoupled Prompt Tuning for Mitigating Spurious Correlation in Vision-Language Models (ICML 2024 · similarity 0.88)
- Analyzing Fine-Grained Alignment and Enhancing Vision Understanding in Multimodal Language Models (NeurIPS 2025 · similarity 0.88)
- Fine-Grained Verifiers: Preference Modeling as Next-token Prediction in Vision-Language Alignment (ICLR 2025 · similarity 0.88)
- With Limited Data for Multimodal Alignment, Let the STRUCTURE Guide You (NeurIPS 2025 · similarity 0.87)
- MmAP: Multi-Modal Alignment Prompt for Cross-Domain Multi-Task Learning (AAAI 2024 · similarity 0.87)
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.
