---
id: 20260822-T0-12
title: "机械断层扫描：面向控制导向可解释性的新测量方法"
title_en: "Mechanistic Tomography: New Method for Interpretability"
url: https://ai.daily.yangsir.net/daily/20260822-T0-12
issue_date: 2026-08-22
publish_date: 2026-08-21T04:00:00.000Z
category: research
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2608.19338
---

# 机械断层扫描：面向控制导向可解释性的新测量方法

一篇新论文提出“机械断层扫描”（Mechanistic Tomography）方法，用于提升大模型的可解释性。该方法旨在通过设计测量来识别模型内部表示的状态、组件效应及相互作用等。该研究将带动量、梯度、Hessian向量积和子集干预等技术结合，为控制模型行为提供更精确的工具。这一方法有望帮助研究人员更好地理解和控制AI系统。

## English Version

**Mechanistic Tomography: New Method for Interpretability**

A new paper introduces 'Mechanistic Tomography,' a method for control-oriented interpretability in large models. The approach involves designing measurements to identify internal states, component effects, and interactions. By combining techniques like patching, gradients, and Hessian-vector products, the method offers more precise tools for controlling model behavior, potentially aiding researchers in understanding and steering AI systems.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2608.19338)

**详情页**：https://ai.daily.yangsir.net/daily/20260822-T0-12

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*