#natural language process...
Apr 2026
Decompose, Look, and Reason: Reinforced Latent Reasoning for VLMs
DLR is proposed, a reinforced latent reasoning framework that dynamically decomposes queries into textual premises, extracts premise-conditioned continuous visual latents, and deduces answers through grounded rationales to enable effective exploration in the latent space.
Mengdan Zhu, Senhao Cheng, Liang Zhao
· arXiv.org · 0 citations