BriefGPT.xyz
Jul, 2024
潜在因果探索:基于数据的因果模型的形式化视角
Latent Causal Probing: A Formal Perspective on Probing with Causal Models of Data
HTML
PDF
Charles Jin
TL;DR
使用结构性因果模型分析了探索语言模型中潜在概念的能力,通过在合成的网格世界导航任务中进行实证研究提供了强有力的证据。
Abstract
As
language models
(LMs) deliver increasing performance on a range of NLP tasks,
probing classifiers
have become an indispensable technique in the effort to better understand their inner workings. A typical setup
→