Research map / Vision-language and multimodal / Multimodal LLMs
HyLaR: Hybrid Latent Reasoning with Decoupled Policy Optimization
Tao Cheng, Shi-Zhe Chen, Hao Zhang, Yixin Qin, Jinwen Luo, Zheng Wei
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 |
|---|---|---|---|
| add latent visual reasoning (vs text-only SFT) (ablation) | HyLaR-SFT · V* | accuracy (%): 69.11 -> 80.63 | Improved |
| add latent visual reasoning (vs text-only SFT) (ablation) | HyLaR-SFT · HRBench-8K | accuracy (%): 61.00 -> 67.19 | Improved |
| add DePO RL stage over SFT | HyLaR-SFT · V* | accuracy (%): 80.63 -> 83.77 | Improved |
| use DePO instead of GRPO (ablation) | HyLaR RL · V* | accuracy (%): 78.53 -> 83.77 | Improved |
| use GRPO RL on top of SFT (ablation) | HyLaR-SFT · V* | accuracy (%): 80.63 -> 78.53 | Hurt |
| use DAPO RL on top of SFT (ablation) | HyLaR-SFT · V* | accuracy (%): 80.63 -> 79.06 | Hurt |
| use VLPO RL on top of SFT (ablation) | HyLaR-SFT · average over 4 benchmarks | accuracy (%): 72.58 -> 72.65 | No clear effect |
| add latent loss weight alpha=0.5 (vs 0) (ablation) | HyLaR-7B · V* | accuracy (%): 80.63 -> 83.77 | Improved |
| raise latent loss weight to 2.0 (vs 0.5) (ablation) | HyLaR-7B · V* | accuracy (%): 83.77 -> 80.63 | Hurt |
| use latent clipping 0.05 instead of 0.2 shared with text (ablation) | HyLaR-7B · V* | accuracy (%): 80.10 -> 83.77 | Improved |
| use no latent clipping (0.0) vs 0.05 (ablation) | HyLaR-7B · V* | accuracy (%): 79.58 -> 83.77 | Improved |
| use vMF instead of Gaussian latent distribution (ablation) | HyLaR RL · SeedBench2Plus | accuracy (%): 67.28 -> 70.32 | Improved |
Closest papers by meaning
- Hybrid Latent Reasoning via Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Thinking on the Fly: Test-Time Reasoning Enhancement via Latent Thought Policy Optimization (ICLR 2026 · similarity 0.88)
- Latent Visual Reasoning (ICLR 2026 · similarity 0.88)
- Why Struggle with Continuous Latents? Interpretable Discrete Latent Reasoning via Rendered Compression (NeurIPS 2026 · similarity 0.88)
- SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows (NeurIPS 2026 · similarity 0.88)
- Markovian Transformers for Informative Language Modeling (ICLR 2026 · 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.
