20260720
This commit is contained in:
62
concepts/contextual-transformation-field.md
Normal file
62
concepts/contextual-transformation-field.md
Normal file
@@ -0,0 +1,62 @@
|
||||
---
|
||||
title: "Contextual Transformation Field (上下文变换场)"
|
||||
created: 2026-07-10
|
||||
updated: 2026-07-10
|
||||
type: concept
|
||||
tags: ["representation-learning", "vector-field", "contextual-representation", "llm-geometry"]
|
||||
sources: ["[[shared-concept-geometry-2026|Hu et al. (ICML 2026)]]"]
|
||||
---
|
||||
|
||||
# Contextual Transformation Field (上下文变换场)
|
||||
|
||||
Hu et al. (ICML 2026) 的核心创新——将 LLM 中概念在上下文间的位移形式化为**向量场**。
|
||||
|
||||
## 定义
|
||||
|
||||
设 τ₀ 为源上下文(如中性模板),τ 为目标上下文。概念 w 的上下文位移:
|
||||
|
||||
```
|
||||
φ^{(m)}(w, τ) = r^{(m)}(w, τ) - r^{(m)}(w, τ₀) ∈ M^{(m)}
|
||||
```
|
||||
|
||||
跨词汇 W 的位移集合构成向量场:
|
||||
|
||||
```
|
||||
Φ_τ^{(m)} : W → M^{(m)}, w ↦ φ^{(m)}(w, τ)
|
||||
```
|
||||
|
||||
由其核矩阵表征:
|
||||
|
||||
```
|
||||
K_Φτ [i, j] = k(φ^{(m)}(w_i, τ), φ^{(m)}(w_j, τ))
|
||||
```
|
||||
|
||||
## 关键发现
|
||||
|
||||
### 场不是均匀的
|
||||
|
||||
- **主导方差比** ρ₁ ∈ [0.073, 0.146]——远未达到单一方向主导
|
||||
- **球面全局方差** V_global ∈ [0.43, 0.75] rad²(均匀分布 ~2.47)——场分散但非随机
|
||||
|
||||
### 方差是语义组织的
|
||||
|
||||
| 词属性 | 与 | 相关性 | p 值 |
|
||||
|--------|-----|--------|------|
|
||||
| 词汇密度 | 位移幅度 r̃ | ρ ∈ [-0.22, -0.32] | < 0.001 |
|
||||
| 具体性 | 方向偏差 σ | ρ ∈ [-0.21, -0.41] | < 0.001 |
|
||||
|
||||
→ 语义邻域密集的词位移更小;更具体的概念方向偏差更小。抽象概念因缺乏感知锚定,在语义框架间被推向更分散的方向。
|
||||
|
||||
### 跨模型共享
|
||||
|
||||
位移关系结构在模型间可迁移:从模型 A 运送位移结构到模型 B,预测保留位移显著高于基线。
|
||||
|
||||
## 对表示工程的意义
|
||||
|
||||
单向量 steering(如 Arditi et al., 2024)将上下文变换场近似为均匀平移 → **丢弃了语义上有意义的结构化残差**。
|
||||
|
||||
## 参考
|
||||
|
||||
- [[shared-concept-geometry-2026|Hu et al. (ICML 2026)]]
|
||||
- [[concept-point-cloud-manifold|Point-Cloud Manifold]]
|
||||
- [[within-concept-between-concept-axes|Two Axes]]
|
||||
Reference in New Issue
Block a user