发表机构
Seoul National University; Konkuk University（首尔大学; 建国大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究提出基于电路的框架，结合训练动态数据估值与机制可解释性，引入SAMS方法实现可控数据生成，在多项选择问答任务中提升了下游性能与校准度。
AI 中文摘要
尽管近期数据合成领域的进展旨在构建高质量数据集，但大多数生成流程仍依赖基于启发式提示的控制。这种黑箱范式难以深入理解单个样本如何与模型的底层学习动态相互作用。为弥合这一差距，我们提出了一种基于电路的框架，将基于训练动态的数据估值与机制可解释性（MI）相结合。具体而言，我们从三个互补的效用轴（可学习性、挑战性、对齐性）来概念化数据质量。首先，我们发现了专门的模型内部电路，这些电路因果性地控制这些效用信号。随后，我们超越启发式提示，转向机制控制，利用这些电路作为可控接口，主动引导生成过程以产生具有目标效用的数据。基于这一能力，我们引入了SAMS（Stage-Aware Mechanistic Scheduling，阶段感知机制调度），该方法根据模型不断变化的优化需求来调度由电路引导的数据。在多项选择问答任务上的实验表明，我们的方法能够生成精确控制且比基于提示的基线更具多样性的数据，持续提升下游性能和校准度。最终，本研究建立了一种用于可解释数据生成的原则性白箱范式，开创了将MI不仅作为分析工具，还作为实用可控接口的应用。
英文摘要
While recent advances in data synthesis aim to curate high-quality datasets, most generation pipelines still rely on heuristic prompt-based control. This black-box paradigm provides limited insight into how individual samples interact with a model's underlying learning dynamics. To bridge this gap, we propose a circuit-grounded framework that connects training-dynamics-based data valuation with mechanistic interpretability (MI). Specifically, we conceptualize data quality along three complementary utility axes, learnability, challenge, and alignment. First, we uncover specialized model-internal circuits that causally govern these utility signals. Then, moving beyond heuristic prompting toward mechanistic control, we leverage these circuits as controllable interfaces, actively steering generation to produce utility-targeted data. Building on this capability, we introduce SAMS (Stage-Aware Mechanistic Scheduling), which schedules circuit-steered data according to the model's evolving optimization needs. Experiments on multiple-choice QA tasks demonstrate that our approach yields precisely controlled data with greater diversity than prompt-based baselines, consistently improving downstream performance and calibration. Ultimately, this work establishes a principled white-box paradigm for interpretable data generation, pioneering the use of MI not just as an analytical tool, but as a practical, controllable interface.
Comments21 pages, 8 figures
