Parameter-Efficient Adapter Tuning for Tabular-Image Multimodal Learning
About
Tabular-image multimodal learning aims to improve predictive modeling by jointly using structured tabular attributes and visual data. Although pretrained encoders provide strong modality-specific representations, full fine-tuning can be computationally expensive, while keeping encoders frozen may limit task-specific adaptation. We propose the Tabular-Image Adapter (TI-Adapter), a modality-specific adapter-based fine-tuning framework for efficient multimodal adaptation. TI-Adapter freezes the pretrained tabular encoder and learns an adapter after the extracted tabular embedding, while adapting the image branch with embedding-level and bottleneck-level adapters instead of full fine-tuning. Experiments on 20 tabular-image datasets show that TI-Adapter achieves competitive or better predictive performance than full fine-tuning while using substantially fewer trainable parameters. Ablation studies further demonstrate the importance of adapter placement for balancing performance and practical efficiency.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Meme Classification | HatefulMemes | Accuracy73.46 | 65 | |
| Classification | PetFinder | Accuracy35.03 | 5 | |
| Classification | glaucoma-smdg | Accuracy87.66 | 5 | |
| Classification | hubmap-hpa | Accuracy69.2 | 5 | |
| Classification | CBIS-DDSM | Accuracy66.29 | 5 | |
| Classification | zooscan-zooplankton | Accuracy92.01 | 5 | |
| Classification | justin-instagram | Accuracy87.84 | 5 | |
| Regression | painting-price | MSE (×10^7)1.54 | 5 | |
| Regression | amazon-bestseller | MSE0.598 | 5 | |
| Regression | Amazon Packages | MSE9.02 | 5 |