Files
myWiki/concepts/loss-landscape.md

37 lines
1.3 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: "Loss Landscape: 神经网络的损失景观"
created: 2026-06-25
updated: 2026-06-25
type: concept
tags: [optimization, deep-learning-theory, loss-surface, mode-connectivity]
sources: ["[[sen-mapping-networks]]"]
---
# Loss Landscape (损失景观)
Loss Landscape 指将神经网络训练目标 L(θ) 视为参数空间 R^P 上的高维曲面时的几何结构。该视角为理解泛化、优化难度和参数空间结构提供了关键洞察。
## 关键发现
### 低维结构
- **Intrinsic Dimension**:深度网络的 objective landscape 的有效内在维度远低于 PLi et al., 2018
- 训练轨迹聚焦在远低于参数空间维度的子空间上
### Mode Connectivity
- SGD 找到的不同局部极小值之间存在**低损路径**Garipov et al., 2018
- 表明有效解分布在连通区域而非孤立点
### 平坦 vs 尖锐极小值
- **平坦极小值**泛化更好Keskar et al., 2017
- Entropy-SGD 等方法显式偏置梯度下降走向宽阔山谷Chaudhari et al., 2019
## 对 Mapping Networks 的意义
Mapping Networks 通过**结构性约束**将搜索空间限制在低维流形上,自然地偏向发现更平坦、更鲁棒的参数解——这是一种通过架构选择实现隐式正则化的方式。
## 参考
- [[intrinsic-dimension]]
- [[manifold-hypothesis]]
- [[weight-manifold-hypothesis]]