透過潛在空間推理！讓潛在視覺推理成為必要

摘要

潛在視覺推理旨在透過隱藏狀態計算來執行多模態推理，而非透過顯式文本思維鏈。然而，視覺資訊存在於潛在狀態中，並不表示模型在產生答案時實際依賴該狀態，尤其當其他以圖像為條件的路徑仍然可用時。我們提出因果視覺遞迴推理（CVRR），其在保留預訓練視覺能力的同時，使遞迴計算成為預測所必需、以圖像為條件的路徑。CVRR 在預訓練視覺語言模型納入圖像後，從問題隱藏狀態初始化遞迴，然後在重新讀取相同固定視覺證據的同時反覆更新此狀態。在解碼前，視覺狀態與原始多模態 KV 快取會被移除，因此僅有最終遞迴狀態將圖像條件資訊帶至答案。在 V^*、MMVP、BLINK 與 MME-RealWorld-Lite 基準測試中，CVRR 在此嚴格介面下維持強勁表現，而相容的潛在推理器即使在同一限制下重新訓練，也無法恢復可比的視覺能力。因果干預進一步顯示，當問題保持固定時，預測仍對遞迴內容敏感，且持續存在的視覺證據會因果性地修正遞迴軌跡。這些結果區分了潛在資訊性與實際用於預測的潛在計算。

English

Latent visual reasoning aims to perform multimodal reasoning through hidden-state computation rather than explicit textual chains of thought. However, visual information being present in a latent state does not imply that the model actually relies on that state when producing its answer, especially when alternative image-conditioned paths remain available. We introduce Causal Visual Recurrent Reasoning (CVRR), which preserves pretrained visual competence while making recurrent computation the required image-conditioned path to prediction. CVRR initializes recurrence from the question hidden state after the pretrained vision-language model has incorporated the image, then repeatedly updates this state while re-reading the same fixed visual evidence. Before decoding, visual states and the original multimodal KV cache are removed so that only the final recurrent state carries image-conditioned information to the answer. Across the V^*, MMVP, BLINK, and MME-RealWorld-Lite benchmarks, CVRR retains strong performance under this strict interface, while compatible latent reasoners fail to recover comparable visual competence even when retrained under the same constraint. Causal interventions further show that predictions remain sensitive to recurrent content when the question is held fixed, and that persistent visual evidence causally revises the recurrent trajectory. These results distinguish latent informativeness from latent computation that is actually used for prediction.