循环、思考与泛化：递归深度变换器中的隐式推理
Loop, Think, & Generalize: Implicit Reasoning in Recurrent-Depth Transformers
AI总结：
本文研究了隐式推理能力，探讨了递归深度变换器在系统泛化和深度扩展中的作用，发现其能有效提升多跳推理能力，但过度递归会导致泛化能力下降。
摘要：
我们研究了隐式推理，即在单次前向传递中结合知识或规则的能力。尽管基于变换器的大语言模型存储了大量事实性知识和规则，但它们往往无法将这些知识组合用于隐式多跳推理，表明在参数知识上的组合泛化能力不足。为了解决这一限制，我们研究了递归深度变换器，它使可以在同一组变换器层上进行迭代计算。我们探讨了隐式推理场景下的两种组合泛化挑战：系统泛化，即结合在训练过程中从未用于组合的知识，以及深度扩展，即从有限的推理深度（例如训练到5跳）扩展到更深层次的组合（例如10跳）。通过受控研究，我们发现虽然普通变换器在两者上都面临困难，但递归深度变换器能够有效实现这种泛化。对于系统泛化，我们发现这种能力通过三个阶段的“grokking”过程出现，从记忆到分布内泛化，最后到系统泛化，这得到了机理分析的支持。对于深度扩展，我们发现通过扩展推理时的递归次数可以解锁超出训练深度的泛化能力，更多的迭代次数能实现更深层次的推理。我们进一步研究了训练策略如何影响扩展能力，提供了指导训练递归深度变换器的建议，并识别了一个关键限制，即过度思考，其中过度的递归会降低预测质量并限制泛化到非常深的组合。
英文摘要：
We study implicit reasoning, i.e. the ability to combine knowledge or rules within a single forward pass. While transformer-based large language models store substantial factual knowledge and rules, they often fail to compose this knowledge for implicit multi-hop reasoning, suggesting a lack of compositional generalization over their parametric knowledge. To address this limitation, we study recurrent-depth transformers, which enables iterative computation over the same transformer layers. We investigate two compositional generalization challenges under the implicit reasoning scenario: systematic generalization, i.e. combining knowledge that is never used for compositions during training, and depth extrapolation, i.e. generalizing from limited reasoning depth (e.g. training on up to 5-hop) to deeper compositions (e.g. 10-hop). Through controlled studies with models trained from scratch, we show that while vanilla transformers struggle with both generalization challenges, recurrent-depth transformers can effectively make such generalization. For systematic generalization, we find that this ability emerges through a three-stage grokking process, transitioning from memorization to in-distribution generalization and finally to systematic generalization, supported by mechanistic analysis. For depth extrapolation, we show that generalization beyond training depth can be unlocked by scaling inference-time recurrence, with more iterations enabling deeper reasoning. We further study how training strategies affect extrapolation, providing guidance on training recurrent-depth transformers, and identify a key limitation, overthinking, where excessive recurrence degrades predictions and limits generalization to very deep compositions.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
