arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.02385cs.LG
收藏

Transformer 学习因式分解表示

Transformers learn factored representations

Simplex, Astera Institute;Sussex Centre for Consciousness Science, University of Sussex;Centre for Complexity Science;Center for Psychedelic Research, Department of Brain Sciences, Imperial College London;Center for Eudaimonia;Human Flourishing, University of Oxford;Beyond Institute for Theoretical Science (BITS)

Adam Shai, Loren Amdahl-Culleton, Casper L. Christensen, Henry R. Bigelow, Fernando E. Rosas, Alexander B. Boyd, Eric A. Alt, Kyle J. Ray, Paul M. Riechers

AI总结:

Transformer通过预训练学习将世界分解为部分,并在正交子空间中表示因素,即使在条件独立性被破坏时仍倾向于保持这种因式分解表示。

摘要:

Transformer 通过下一个token预测进行预训练,学习将世界分解为部分,并在残差流的正交子空间中表示这些因素。我们正式提出了两个表示假设:(1) 所有因素的产品空间中的表示,其维度随部分数量呈指数增长,或(2) 正交子空间中的因式分解表示,其维度线性增长。当因素条件独立时,因式分解表示是无损的,但其他情况下会牺牲预测保真度,从而在维度效率和准确性之间产生权衡。我们为每种情况推导了关于激活几何结构的精确预测,包括子空间的数量、维度性以及其中上下文嵌入的排列。我们测试了这些假设,使用具有已知潜在结构的合成过程训练的Transformer。当因素条件独立时,模型学习因式分解表示,即使在噪声或隐藏依赖性破坏条件独立性时,模型在训练初期仍倾向于保持这种表示,反映出对分解的归纳偏置,即使牺牲保真度。这为为什么Transformer将世界分解为部分提供了系统性的解释,并表明即使在训练于复杂数据的模型中,可解释的低维结构也可能持续存在。

英文摘要:

Transformers pretrained via next token prediction learn to factor their world into parts, representing these factors in orthogonal subspaces of the residual stream. We formalize two representational hypotheses: (1) a representation in the product space of all factors, whose dimension grows exponentially with the number of parts, or (2) a factored representation in orthogonal subspaces, whose dimension grows linearly. The factored representation is lossless when factors are conditionally independent, but sacrifices predictive fidelity otherwise, creating a tradeoff between dimensional efficiency and accuracy. We derive precise predictions about the geometric structure of activations for each, including the number of subspaces, their dimensionality, and the arrangement of context embeddings within them. We test between these hypotheses on transformers trained on synthetic processes with known latent structure. Models learn factored representations when factors are conditionally independent, and continue to favor them early in training even when noise or hidden dependencies undermine conditional independence, reflecting an inductive bias toward factoring at the cost of fidelity. This provides a principled explanation for why transformers decompose the world into parts, and suggests that interpretable low dimensional structure may persist even in models trained on complex data.

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。

↑