Feed Entities Podcasts Deep Dives Developers About

[KongchangAI]

Real-time AI knowledge base. Every claim is traceable to its source, with data freshness under 24 hours.

Powered by 240+ first-hand AI sources

Knowledge

Feed
Hot Entities
Podcasts

Developers

CLI
MCP Server
REST API
About

© 2026 KongchangAI. The freshest, most accurate AI knowledge

Unverified60% confidenceFactExact time

迁移学习中通常有两种策略：冻结预训练模型大部分层只训练分类头（Feature Extraction），或用预训练权重初始化后以较小学习率微调所有层（Fine-Tuning）

2

Sources

60%

Confidence

Medium-term (~90 days)

Relevance

7/2/2026

First Seen

Valid until: 9/30/2026

Sources

PyTorch图像分类实战：数据预处理到模型训练完整流程

bilibili程序员是魔法师AI-bx6/17/2026

Related Claims

UnverifiedTransfer learning has two main strategies: Feature Extraction (freezing most layers and only training the final classification head) and Fine-Tuning (initializing with pretrained weights and training all layers with a smaller learning rate).81% similar UnverifiedWarmup策略在训练初期使用极小学习率，待参数进入合理范围后再提升，被大型语言模型训练广泛采用78% similar Unverified小版本号跃迁通常对应后训练（Post-Training）阶段的优化，而非预训练的重新训练77% similar Unverified不同模型在抗诱导能力上的差异很大程度上源于后训练阶段（post-training）的策略差异74% similar UnverifiedTransition的训练计划能精确到配速区间、心率目标和训练目的，而非简单的距离指令74% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/45217

API

curl https://kongchang.com/api/v1/knowledge/claims/45217

MCP

get_claim(id=45217)