20260601
This commit is contained in:
42
concepts/data-replay.md
Normal file
42
concepts/data-replay.md
Normal file
@@ -0,0 +1,42 @@
|
||||
---
|
||||
title: "数据回放 (Data Replay)"
|
||||
created: 2026-05-21
|
||||
type: concept
|
||||
tags: ["continual-learning", "knowledge-retention", "catastrophic-forgetting"]
|
||||
sources: ["[[when-large-multimodal-models-confront-evolving-knowledge]]"]
|
||||
---
|
||||
|
||||
# 数据回放 (Data Replay)
|
||||
|
||||
## 定义
|
||||
|
||||
数据回放是缓解[[capability-degradation|能力退化]]的一种**直接排练策略**,通过将**旧预训练数据**与新注入数据混合进行微调,强制模型"复习旧知"。
|
||||
|
||||
## 实现
|
||||
|
||||
在 MMEVOKE 论文中:
|
||||
- Replay + Full-FT:随机抽样 10%(MMEVOKE 数据量大小)的旧预训练数据,与新注入数据混合,使用 Full-FT
|
||||
- Replay + LoRA:同上,使用 LoRA 策略
|
||||
|
||||
## 效果(LLaVA-v1.5)
|
||||
|
||||
- Replay + Full-FT:缓解退化,**排名第 3**
|
||||
- Replay + LoRA:**排名第 1**,在 MMMU 和 MathVision 上超过 Vanilla 分别 +1.75% 和 +2.20%
|
||||
|
||||
## 机制
|
||||
|
||||
通过重新暴露于旧数据,**重新激活旧知识网络**,防止新知识覆盖已有参数。
|
||||
|
||||
## 与 MoELoRA 的比较
|
||||
|
||||
| 策略 | Replay | MoELoRA |
|
||||
|------|--------|---------|
|
||||
| 机制 | 直接排练旧数据 | 结构性隔离新知识 |
|
||||
| 优势 | 效果最佳 | 无需存储旧数据 |
|
||||
| 劣势 | 需要旧数据存储 | 需修改模型架构 |
|
||||
|
||||
## 参见
|
||||
|
||||
- [[moe-lora|MoELoRA]]
|
||||
- [[knowledge-retention|知识保留]]
|
||||
- [[continual-learning|持续学习]]
|
||||
Reference in New Issue
Block a user