Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Revisit Multimodal Meta-Learning through the Lens of Multi-Task Learning
Milad Abdollahzadeh, Touba Malekzadeh, Ngai-man Cheung
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 KML modulation to MMAML | MMAML · Multimodal 2Mode 5-shot | accuracy: 73.52 -> 77.02 | Improved |
| add KML modulation to MProtoNet | MProtoNet · Multimodal 5Mode 5-shot | accuracy: 59.95 -> 64.91 | Improved |
| add KML modulation to MMAML | MMAML · Multimodal 2Mode-dagger 5-shot | accuracy: 52.02 -> 53.72 | Improved |
| replace MAML with ProtoNet as meta-learner in MMAML | MMAML · Multimodal 2Mode-dagger 1-shot | accuracy: 39.11 -> 43.57 | Improved |
| train one multimodal ProtoNet instead of separate shared model | Multi-ProtoNet vs ProtoNet · Multimodal 2Mode-dagger 5-shot | accuracy: 57.70 -> 56.73 | Hurt |
| modulate all layers (no shared layers) in multimodal setting (ablation) | MProtoNet+KML · Multimodal 2Mode-dagger 1-shot | accuracy: 43.05 -> 44.40 | Improved |
| share first layer in multimodal setting (ablation) | MProtoNet+KML · Multimodal 2Mode-dagger 1-shot | accuracy: 44.40 -> 44.18 | Hurt |
| share first two layers in unimodal KML (ablation) | KML-ProtoNet · mini-ImageNet 1-shot | accuracy: 51.55 -> 54.10 | Improved |
| share first, second and third layers in unimodal KML (ablation) | KML-ProtoNet · mini-ImageNet 1-shot | accuracy: 54.10 -> 52.83 | Hurt |
| share only the first layer in unimodal KML (ablation) | KML-ProtoNet · mini-ImageNet 5-shot | accuracy: 67.18 -> 68.07 | Improved |
| unimodal KML with two shared layers (ablation) | KML-ProtoNet · tiered-ImageNet 1-shot | accuracy: 53.01 -> 54.67 | Improved |
| add KML to MMAML for RL | MMAML · Point Mass 2D 2 modes | cumulative reward: -136 -> -121 | Improved |
Closest papers by meaning
- Multimodal Model-Agnostic Meta-Learning via Task-Aware Modulation (NeurIPS 2019 · similarity 0.89)
- $\pi$-Tuning: Transferring Multimodal Foundation Models with Optimal Multi-task Interpolation (ICML 2023 · similarity 0.88)
- Meta-Learning of Structured Task Distributions in Humans and Machines (ICLR 2021 · similarity 0.88)
- Advances and Challenges in Meta-Learning: A Technical Review (TPAMI 2024 · similarity 0.88)
- MMPareto: Boosting Multimodal Learning with Innocent Unimodal Assistance (ICML 2024 · similarity 0.86)
- Multimodal Representation Learning by Alternating Unimodal Adaptation (CVPR 2024 · similarity 0.86)
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.
