Transformer注意力头像博弈玩家，训练方法导致冗余与幻觉，新理论可优化。
Multi-Head Attention Is a Multi-Player Game
- 将注意力头视为博弈参与者，用潜在博弈建模其竞争协作关系。
- 提出价格悖论上限Γ(G)，可预测幻觉概率和头冗余度，实验验证显著降低幻觉。
- 设计GAME-LoRA正则化方法，无需牺牲性能即减少幻觉18%（平均8%）。
现代Transformer注意力机制本质上是多智能体系统——注意力头之间既竞争又协作，但训练时却视作单一优化器。本文揭示这一差距：交叉熵训练在头间诱导隐式潜在博弈，梯度下降收敛至纳什均衡，可能因未定价的外部性（如冗余、相关错误）导致效率无界损失。核心结果证明，价格悖论（PoA）由头交互矩阵非对角线质量Γ(G)界定，该矩阵刻画权重与梯度耦合。在温和平滑假设下，我们证明了‘额外幻觉概率’与‘额外头冗余’均随PoA增长，统一两种失效模式为单一机制。该边界具有指导意义：减小Γ(G)的正则化可严格收紧PoA。我们据此提出GAME-LoRA，融合Barlow Twins去相关与对数行列式协调压力。实验验证理论：Γ(G)可预测幻觉（p<0.05），涌现同盟呈现选择性协调；GAME-LoRA实现最高18%的幻觉降低（平均8%），且无知识损失——达成帕累托改进，这是忽略博弈结构的方法无法实现的。
原文摘要 · Abstract (English)
Modern transformer attention is internally multi-agent -- heads compete and coordinate -- yet we train it as if it were a monolithic optimizer. We formalize this gap: cross-entropy training induces an implicit potential game among heads, and gradient descent converges to Nash equilibria with potentially unbounded inefficiency due to unpriced externalities (redundancy, correlated errors). Our main result bounds the Price of Anarchy by $Γ(G)$, the off-diagonal mass of a head interaction matrix capturing weight and gradient coupling. Under mild smoothness assumptions, we prove that both \emph{excess hallucination probability} and \emph{excess head redundancy} scale with PoA, unifying two distinct failure modes into a single mechanism. The bound is prescriptive: regularization that reduces $Γ(G)$ provably tightens PoA. We instantiate this as GAME-LoRA, combining Barlow Twins decorrelation with log-determinant coordination pressure. Experiments validate the theory: $Γ(G)$ predicts hallucination ($p{<}0.05$), emergent coalitions exhibit selective coordination, and GAME-LoRA achieves up to 18\% hallucination reduction (8\% average) with no knowledge degradation -- a Pareto improvement inaccessible to methods ignoring the game structure.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。