Research map / Architectures and efficiency / Attention and transformers
UniT: Multimodal Multitask Learning With a Unified Transformer
Ronghang Hu, Amanpreet Singh
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 |
|---|---|---|---|
| joint training with separate decoders instead of single-task training | UniT · VQAv2 (COCO+VG+VQAv2) | accuracy: 66.38 -> 68.84 | Improved |
| share one decoder across tasks instead of separate decoders | UniT · COCO det. | mAP: 40.8 -> 37.2 | Hurt |
| initialize shared-decoder joint model from COCO single-task model | UniT shared decoder · COCO det. | mAP: 37.2 -> 40.8 | Improved |
| add VG detection and COCO detection together to VQAv2 training | UniT shared (COCO init.) · VG det. | mAP: 3.83 -> 4.53 | Improved |
| use VG detection instead of COCO detection as the detection partner for VQAv2 | UniT shared (COCO init.) · VQAv2 | accuracy: 66.88 -> 68.49 | Improved |
| joint multi-task training on 8 datasets instead of single-task training | UniT shared (COCO init.) · SNLI-VE | accuracy: 70.52 -> 73.16 | Improved |
| joint multi-task training on 8 datasets instead of single-task training | UniT shared (COCO init.) · COCO det. | mAP: 40.6 -> 39.0 | Hurt |
| per-task finetuning of shared model | UniT · COCO det. | mAP: 39.0 -> 42.3 | Improved |
| per-task finetuning of shared model | UniT · SST-2 | accuracy: 89.29 -> 88.06 | Hurt |
| increase decoder layers to 8 (ablation) | UniT · COCO det. (ablation) | mAP: 38.79 -> 40.13 | Improved |
| double decoder layers to 12 with 400k iterations (ablation) | UniT · COCO det. (ablation) | mAP: 38.79 -> 39.02 | No clear effect |
| use larger decoder hidden size 768 instead of 256 (ablation) | UniT · COCO det. (ablation) | mAP: 36.32 -> 38.79 | Improved |
Closest papers by meaning
- OmniVec2 - A Novel Transformer based Network for Large Scale Multimodal and Multitask Learning (CVPR 2024 · similarity 0.86)
- UniTS: A Unified Multi-Task Time Series Model (NeurIPS 2024 · similarity 0.86)
- Emergence and Effectiveness of Task Vectors in In-Context Learning: An Encoder Decoder Perspective (ICML 2025 · similarity 0.86)
- L-Man: A Large Multi-modal Model Unifying Human-centric Tasks (AAAI 2025 · similarity 0.85)
- UNIT: Unifying Image and Text Recognition in One Vision Encoder (NeurIPS 2024 · similarity 0.85)
- HaploVL: A Single-Transformer Baseline for Multi-Modal Understanding (ICML 2025 · similarity 0.85)
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.
