Research map / Vision-language and multimodal / Multimodal LLMs
StochasT: Learning with Stochastic Turn Depth for Visual Instruction Tuning
Yuan Qing, Chengzhi Mao, Boqing Gong
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 StochasT instead of MultiT training | LLaVA-1.5-7B · Avg of 4 VIT datasets (MultiT eval) | accuracy: 68.52 -> 70.63 | Improved |
| use StochasT instead of MultiT training | Qwen2.5-VL-3B · Avg of 4 VIT datasets (MultiT eval) | accuracy: 77.39 -> 80.16 | Improved |
| use StochasT instead of MultiT training | LLaVA-1.5-7B · Avg of 4 VIT datasets (SingleT eval) | accuracy: 61.51 -> 67.16 | Improved |
| use StochasT instead of MultiT training | Qwen2.5-VL-3B · Avg of 4 VIT datasets (SingleT eval) | accuracy: 71.20 -> 77.28 | Improved |
| use StochasT instead of MultiT training | LLaVA-1.5-7B · MMDU | score: 12.68 -> 13.10 | Improved |
| use StochasT instead of MultiT training | Qwen2.5-VL-3B · MMDU | score: 57.90 -> 59.80 | Improved |
| use StochasT instead of MultiT training | LLaVA-1.5-7B · Avg of 4 VIT datasets (BLS eval) | CRA / CRA+: CRA 61.82 -> 67.69; CRA+ 41.19 -> 50.89 | Improved |
| use StochasT instead of SingleT training | LLaVA-1.5-7B · Avg of 4 VIT datasets (BLS eval) | CRA+: 53.15 -> 50.89 | Hurt |
| use StochasT instead of SingleT training | Qwen2.5-VL-3B · Avg of 4 VIT datasets (SingleT eval) | accuracy: 76.71 -> 77.28 | No clear effect |
| use Beta(2,2) instead of Beta(0.5,0.5) for dropout rates (ablation) | StochasT · CoralVQA + iNat-Plant avg | CRA+: 69.05 -> 73.30 | Improved |
| use Beta(5,1) instead of Beta(1,5) for dropout rates (ablation) | StochasT · CoralVQA + iNat-Plant avg | CRA+: 69.81 -> 72.69 | Improved |
| replace StochasT with random-fixed cutoff at matched expected depth (ablation) | Qwen2.5-VL-3B · iNat-Plant + PathVQA avg | CRA: 79.24 -> 70.61 | Hurt |
Closest papers by meaning
- VILA: On Pre-training for Visual Language Models (CVPR 2024 · similarity 0.88)
- Enhancing Large Vision Language Models with Self-Training on Image Comprehension (NeurIPS 2024 · similarity 0.88)
- A Comprehensive Information-Decomposition Analysis of Large Vision-Language Models (ICLR 2026 · similarity 0.88)
- Learning to Instruct for Visual Instruction Tuning (NeurIPS 2025 · similarity 0.88)
- Prismatic VLMs: Investigating the Design Space of Visually-Conditioned Language Models (ICML 2024 · similarity 0.88)
- MMDU: A Multi-Turn Multi-Image Dialog Understanding Benchmark and Instruction-Tuning Dataset for LVLMs (NeurIPS 2024 · 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.
