← 研踪 PaperTrailACL Anthology · ACL 2026:acl:2026.acl-long.809 · 记录 #1
Tracing Logit Trajectories Across Layer Depth: Dataset-Level Explainability for Language Models
来源日期（仅到月份） 2026-07
来源记录日期 2026-07
雷达观测 2026-11-01 21:58
摘要
Sentence-level explanations can miss the bigger picture of how a black-box model behaves across data, which matters most for complex criteria like safety that cannot be defined by a single rule. We trace Logit-Trajectory, which tracks adjacent-layer logit updates as vectors and aggregates them into a reproducible dataset-level trajectory pattern, enabling depth-wise explainability through signals such as coherence and angular rotation. Across 6 languages and 5 NLP tasks, we show these trajectory summaries reveal consistent depth-wise patterns that divergence- and similarity-based baselines often wash out due to scalarization. As a case study where dataset-level intermediate decision structure matters, we evaluate safety classification, reporting both trajectory-level visual separability and classification performance.
论文发现出处
这些页面提供论文线索，不代表同行评审或质量认证。
观测到的版本
平台记录号代表本站观测到的元数据变化，不是作者修订号。来源日期仅到月份或年份时，按该月或该年首日存档。
记录 #1 · 2026-07 · 最新观测版本
Tracing Logit Trajectories Across Layer Depth: Dataset-Level Explainability for Language Models
Sentence-level explanations can miss the bigger picture of how a black-box model behaves across data, which matters most for complex criteria like safety that cannot be defined by a single rule. We trace Logit-Trajectory, which tracks adjacent-layer logit updates as vectors and aggregates them into a reproducible dataset-level trajectory pattern, enabling depth-wise explainability through signals such as coherence and angular rotation. Across 6 languages and 5 NLP tasks, we show these trajectory summaries reveal consistent depth-wise patterns that divergence- and similarity-based baselines often wash out due to scalarization. As a case study where dataset-level intermediate decision structure matters, we evaluate safety classification, reporting both trajectory-level visual separability and classification performance.
查看此版本解读 →查看此版本原文 ↗