Research map / Vision-language and multimodal / Human and embodied modeling
Benchmarking Multimodal AutoML for Tabular Data with Text Fields
Xingjian Shi, Jonas Mueller, Nick Erickson, Mu Li, Alex Smola
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 |
|---|---|---|---|
| switch text backbone from RoBERTa to ELECTRA (ablation) | Text-Net · 18-dataset multimodal benchmark, average | avg score: 0.572 -> 0.605 | Improved |
| add exponential layer-wise learning rate decay (tau=0.8) (ablation) | ELECTRA · 18-dataset multimodal benchmark, average | avg score: 0.605 -> 0.632 | Improved |
| add averaging of top-3 checkpoints (ablation) | ELECTRA + decay · 18-dataset multimodal benchmark, average | avg score: 0.632 -> 0.635 | Improved |
| Fuse-Late instead of Text-Net (add numeric/categorical inputs) (ablation) | Multimodal-Net · 18-dataset multimodal benchmark, average | avg score: 0.635 -> 0.664 | Improved |
| Fuse-Late instead of Fuse-Early (ablation) | Multimodal-Net · 18-dataset multimodal benchmark, average | avg score: 0.658 -> 0.664 | Improved |
| Fuse-Late instead of All-Text (ablation) | Multimodal-Net · 18-dataset multimodal benchmark, average | avg score: 0.662 -> 0.664 | Improved |
| stack ensemble Multimodal-Net with tabular models instead of weighted ensemble (ablation) | Multimodal-Net · 18-dataset multimodal benchmark, average | avg score: 0.676 -> 0.684 | Improved |
| use fine-tuned text embeddings instead of pretrained (Text-Embedding vs Pre-Embedding) (ablation) | tabular models · 18-dataset multimodal benchmark, average | avg score: 0.593 -> 0.638 | Improved |
| contextualize text embeddings with numeric/categorical info (Multimodal-Embedding vs Text-Embedding) (ablation) | tabular models · 18-dataset multimodal benchmark, average | avg score: 0.638 -> 0.644 | Improved |
| ensemble Multimodal-Net with tabular models instead of using Transformer as embedding (ablation) | Stack-Ensemble · 18-dataset multimodal benchmark, average | avg score: 0.644 -> 0.684 | Improved |
| use text features instead of ignoring them | Stack-Ensemble vs AG-Stack · mercari | score: 0.098 vs 0.605 | Improved |
| use text features instead of ignoring them | Stack-Ensemble vs AG-Stack · kick | score: 0.670 vs 0.797 | Improved |
Closest papers by meaning
- Large Scale Transfer Learning for Tabular Data via Language Modeling (NeurIPS 2024 · similarity 0.89)
- MulTaBench: Benchmarking Multimodal Tabular Learning with Text and Image (NeurIPS 2026 · similarity 0.89)
- STRABLE: Benchmarking Tabular Machine Learning with Strings (NeurIPS 2026 · similarity 0.88)
- UniTabE: A Universal Pretraining Protocol for Tabular Foundation Model in Data Science (ICLR 2024 · similarity 0.88)
- TabDPT: Scaling Tabular Foundation Models on Real Data (NeurIPS 2025 · similarity 0.88)
- TabArena: A Living Benchmark for Machine Learning on Tabular Data (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.
