Files
myWiki/concepts/mapping-theorem.md

50 lines
1.8 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

---
title: "Mapping Theorem: 参数空间的低维映射存在性定理"
created: 2026-06-25
updated: 2026-06-25
type: concept
tags: [theorem, manifold-learning, mapping-networks, existence-proof]
sources: ["[[sen-mapping-networks]]"]
---
# Mapping Theorem (映射定理)
Mapping Theorem 是 [[sen-mapping-networks|Mapping Networks]] 的理论基石,证明了**从低维隐空间到高维参数空间的光滑映射的存在性**,且该映射可在损失函数上任意逼近最优参数。
## 前提条件
1. **A1: 参数光滑性** — θ → f_θ(x) 是 L_θ-Lipschitz 的(对每个 x
2. **A2: 损失 Lipschitz** — L(·, y) 是 L_-Lipschitz 的
3. **A3: 局部可逼近性** — M_θ 是 C² 流形,有界曲率
4. **Weight-Manifold Hypothesis** — θ* 位于 C² 嵌入流形 M_θ ⊂ R^P 上
## 定理陈述
对任意 ε > 0满足 ε ≤ L_ L_θ r存在
- δ > 0
- d ≥ d*(其中 d* = dim(M_θ)
- C² 映射 g: R^d → R^P
- 隐向量 z* ∈ R^d
使得:
$$\|g(z^*) - \theta^*\| \leq \delta, \quad |L(g(z^*)) - L(\theta^*)| \leq \varepsilon$$
## 证明概要
1. 由 Weight-Manifold Hypothesis∃ C² 微分同胚 φ: U → V ⊂ M_θφ(0) = θ*
2. 构造全局映射 g(u) = ψ(u)φ(u) + (1 ψ(u))θ*smooth bump function 拼接)
3. 由连续性,选 z* ∈ B(0, η) ∩ U满足 ‖g(z*) θ*‖ < δ
4. Lipschitz 条件|L(g(z*)) L(θ*)| L_ L_θ · δ = ε
## 实际意义
该定理提供了**架构设计的正确性保证**如果映射网络架构满足定理条件 [[solvability-theorem|Solvability Theorem]] 所示则理论上存在隐向量可生成与完整训练等效的参数
## 参考
- [[weight-manifold-hypothesis]]
- [[solvability-theorem]]
- [[mapping-loss]]
- Sen & Mukherjee, "Mapping Networks", arXiv:2602.19134, Section 2.1