生成式遞迴推理

摘要

未來的神經推理系統應如何實現擴展計算？遞歸推理模型（RRMs）透過共享轉換函數進行迭代潛在狀態精煉，為自迴歸序列擴展提供了有前景的替代方案。然而現有的RRMs大多為確定性模型，遵循單一潛在軌跡並收斂至單一預測。我們引入了生成式遞歸推理模型（GRAM），這是一個將遞歸潛在推理轉化為機率性多軌跡計算的框架。GRAM將推理建模為隨機潛在軌跡，從而支持多種假設、替代解題策略，並透過遞歸深度和平行軌跡取樣實現推理時擴展。這產生了一個潛在變數生成模型，支援透過p_θ(y|x)進行條件推理，以及透過固定或無輸入的p_θ(x)進行無條件生成。透過攤銷變分推斷進行訓練，GRAM在結構化推理與多解約束滿足任務上勝過確定的遞迴與遞歸基線，同時展現無條件生成的能力。https://ahn-ml.github.io/gram-website

English

How should future neural reasoning systems implement extended computation? Recursive Reasoning Models (RRMs) offer a promising alternative to autoregressive sequence extension by performing iterative latent-state refinement with shared transition functions. Yet existing RRMs are largely deterministic, following a single latent trajectory and converging to a single prediction. We introduce Generative Recursive reAsoning Models (GRAM), a framework that turns recursive latent reasoning into probabilistic multi-trajectory computation. GRAM models reasoning as a stochastic latent trajectory, enabling multiple hypotheses, alternative solution strategies, and inference-time scaling through both recursive depth and parallel trajectory sampling. This yields a latent-variable generative model supporting conditional reasoning via p_θ(y mid x) and, with fixed or absent inputs, unconditional generation via p_θ(x). Trained with amortized variational inference, GRAM improves over deterministic recurrent and recursive baselines on structured reasoning and multi-solution constraint satisfaction tasks, while demonstrating an unconditional generation capability. https://ahn-ml.github.io/gram-website