Research map / Vision-language and multimodal / Cross-modal learning
Understanding Multimodal Learning: A Loss Landscape Smoothness Perspective
Jae-Jun Lee, Sung Whan Yoon
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 |
|---|---|---|---|
| use stochastic modality pairing (DML) instead of fixed pairing | SML · Kinetics-Sounds | accuracy: 62.95 -> 65.89 | Improved |
| use stochastic modality pairing (DML) instead of fixed pairing | SML · AVMNIST | accuracy: 70.30 -> 71.69 | Improved |
| use stochastic modality pairing (DML) instead of fixed pairing | SML · CREMA-D | accuracy: 64.95 -> 65.53 | Improved |
| use stochastic modality pairing (DML) instead of fixed pairing | SML · UPMC-Food101 | accuracy: 91.48 -> 92.84 | Improved |
| add multimodal fusion (SML) over the best unimodal model | [A] unimodal · Kinetics-Sounds | accuracy: 51.95 -> 62.95 | Improved |
| add data augmentation (jitter, crop, flip) | DML · UPMC-Food101 | accuracy: 92.84 -> 94.93 | Improved |
| add data augmentation (jitter, crop, flip) | DML · CREMA-D | accuracy: 65.53 -> 67.39 | Improved |
| add data augmentation (jitter, crop, flip) | SML · Kinetics-Sounds | accuracy: 62.95 -> 64.51 | Improved |
| add Stochastic Weight Averaging | SML · AVMNIST | accuracy: 70.30 -> 69.24 | Hurt |
| add Sharpness-Aware Minimization | SML · CREMA-D | accuracy: 64.95 -> 65.59 | Mixed |
| add label smoothing | SML · UPMC-Food101 | accuracy: 91.48 -> 91.79 | No clear effect |
| use DML instead of SML in contrastive training | CLIP loss on Flickr-30k · Flickr-30k Text-to-Image | R@1: 74.19 -> 76.10 | Improved |
Closest papers by meaning
- Fusion or Confusion? Multimodal Complexity Is Not All You Need (NeurIPS 2026 · similarity 0.90)
- MMPareto: Boosting Multimodal Learning with Innocent Unimodal Assistance (ICML 2024 · similarity 0.89)
- Asymmetric Reinforcing Against Multi-Modal Representation Bias (AAAI 2025 · similarity 0.89)
- Modality-Aware SAM: Sharpness-Aware-Minimization Driven Gradient Modulation for Harmonized Multimodal Learning (NeurIPS 2025 · similarity 0.89)
- Multimodal Representation Learning by Alternating Unimodal Adaptation (CVPR 2024 · similarity 0.88)
- Balancing Multimodal Training Through Game-Theoretic Regularization (NeurIPS 2025 · similarity 0.88)
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.
