Research map / Vision-language and multimodal / Multimodal LLMs
Thinking in Latent Space: Progressive Multimodal Simplification for Visual Reasoning
Yuesen Tang, Yiming Yang, Tengfei Bao, Yu Tong
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a progressive multimodal simplification approach that moves MLLM visual reasoning from explicit textual CoT into latent space (abstract only available).
Problem
Existing MLLM visual reasoning relies heavily on explicit textual chain-of-thought, causing information loss, unstable perception-reasoning interaction, and high computational cost.
Approach
Argues effective visual reasoning emerges from a dynamic process of progressive simplification and performs reasoning in latent space rather than through explicit textual steps; the abstract is truncated so specifics are unknown.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the Latent Transition Kernel (both phases) (ablation) | LDPVR · MMVP | accuracy: 66.5 -> 72.4 | Improved |
| add the Latent Transition Kernel (both phases) (ablation) | LDPVR · MM-Math | accuracy: 40.2 -> 43.7 | Improved |
| add the Latent Transition Kernel (both phases) (ablation) | LDPVR · ScienceQA | accuracy: 48.3 -> 56.1 | Improved |
| add Semantic Intent Anchoring (Phase 1) (ablation) | LDPVR · MMVP | accuracy: 68.0 -> 72.4 | Improved |
| add Semantic Intent Anchoring (Phase 1) (ablation) | LDPVR · MM-Math | accuracy: 42.7 -> 43.7 | Improved |
| add Semantic Intent Anchoring (Phase 1) (ablation) | LDPVR · ScienceQA | accuracy: 53.9 -> 56.1 | Improved |
| add Progressive Latent Grounding (Phase 2) (ablation) | LDPVR · MMVP | accuracy: 65.5 -> 72.4 | Improved |
| add Progressive Latent Grounding (Phase 2) (ablation) | LDPVR · MM-Math | accuracy: 41.9 -> 43.7 | Improved |
| add Progressive Latent Grounding (Phase 2) (ablation) | LDPVR · ScienceQA | accuracy: 51.8 -> 56.1 | Improved |
| progress from latent stage 1 to stage 3 (adds distillation and GRPO) (ablation) | LDPVR · MMVP | accuracy: 66.4 -> 72.4 | Improved |
| add the final GRPO stage after stage 2 (ablation) | LDPVR · ScienceQA | accuracy: 53.4 -> 56.1 | Improved |
| early latent stage 1 versus plain SFT+GRPO baseline (ablation) | LDPVR · MMVP | accuracy: 68.9 -> 66.4 | Hurt |
Limitations
- Only the truncated abstract was available; method details and results are unknown.
Experimental setup
Builds on: chain-of-thought prompting
Closest papers by meaning
- SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows (NeurIPS 2026 · similarity 0.94)
- Latent Visual Reasoning (ICLR 2026 · similarity 0.94)
- Leveraging Latent Visual Reasoning in Silence (NeurIPS 2026 · similarity 0.94)
- Monet: Reasoning in Latent Visual Space Beyond Image and Language (CVPR 2026 · similarity 0.93)
- AStar: Boosting Multimodal Reasoning with Automated Structured Thinking (AAAI 2026 · similarity 0.93)
- ThinkMorph: Emergent Properties in Multimodal Interleaved Chain-of-Thought Reasoning (ICLR 2026 · similarity 0.93)
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.
