发表机构
OATML; University of Oxford; UBC; Google DeepMind（OATML; 牛津大学; 英属哥伦比亚大学; 谷歌DeepMind）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本文提出统一视角，通过token与层级的交互分解，揭示持续学习中数据归因、遗忘与可塑性损失同源于演化动力学，并据此实现数据选择、干扰控制及可学习性诊断。
AI 中文摘要
现代语言模型很可能在其生命周期内被持续更新，而非仅训练一次后便冻结。因此，每次更新都参与一个循环往复的过程：决定从哪些经验中学习，理解该更新改变了什么，并保持从后续经验中学习的能力。我们表明，这些挑战受制于同一不断演化的更新-行为交互作用。我们推导出从某个token学习如何改变另一token预测的token级和层级的分解。通过分离softmax力、共享读出几何和残差连接，该方法揭示了两个交互通道，并产生一种可前向计算的近似。随时间追踪这一交互作用，揭示了持续适应的统一图景。正交互识别有用经验；负交互产生集中碰撞或累积侵蚀；在更长的时间跨度上，更新重塑了介导未来学习信号的共享几何结构，降低其传输效率。这些预测带来了有效的数据选择、针对干扰的机制特定控制，以及一种基于读出的未来可学习性诊断方法，其退化预测了恢复读出的益处。跨模型和训练机制，相同的局部交互因此既解释了更新现在改变了什么，也解释了今天的学习如何改变明天可学的内容。这一视角将数据归因、遗忘和可塑性损失视为同一演化学习动力学的不同状态。
英文摘要
Modern language models are likely to be updated throughout their lifetime rather than trained once and frozen. Each update therefore participates in a recurring cycle: decide which experience to learn from, understand what that update changes, and remain capable of learning from what comes next. We show that these challenges are governed by the same evolving update--behavior interaction. We derive a token- and layer-wise decomposition of how learning from one token changes another prediction. By separating the softmax force, shared readout geometry, and residual connections, it exposes two interaction channels and yields a forward-computable approximation. Following this interaction through time reveals a unified picture of continual adaptation. Positive interaction identifies useful experience; negative interaction produces either concentrated collision or accumulated erosion; over longer horizons, updates reshape the shared geometry mediating future learning signals, reducing their transmission. These predictions lead to effective data selection, mechanism-specific controls for interference, and a readout-based diagnostic of future learnability whose degradation predicts the benefit of restoring the readout. Across models and training regimes, the same local interaction thus explains both what an update changes now and how learning today changes what can be learned tomorrow. This view connects data attribution, forgetting, and plasticity loss as distinct regimes of the same evolving learning dynamics.
Comments45 pages
