Course Code: clmusftm
Duration: 14 hours
Prerequisites:
  • 了解機器學習工作流程與神經網絡架構
  • 具備模型微調與部署管道的經驗
  • 熟悉數據版本控制與模型生命週期管理

目標受眾

  • AI維護工程師
  • MLOps工程師
  • 負責模型生命週期連續性的機器學習從業者
Overview:

持續學習是一套策略，使機器學習模型能夠逐步更新並隨著時間適應新數據。

這門由講師指導的培訓（線上或線下）面向高級AI維護工程師和MLOps專業人士，他們希望為已部署且微調的模型實施穩健的持續學習管道和有效的更新策略。

在本培訓結束時，參與者將能夠：

  • 設計並實施已部署模型的持續學習工作流程。
  • 通過適當的訓練和記憶管理來減輕災難性遺忘。
  • 根據模型漂移或數據變化自動化監控和更新觸發機制。
  • 將模型更新策略整合到現有的CI/CD和MLOps管道中。

課程形式

  • 互動式講座和討論。
  • 大量練習和實踐。
  • 在實時實驗室環境中進行實際操作。

課程定制選項

  • 如需為本課程定制培訓，請聯繫我們進行安排。
Course Outline:

持續學習簡介

  • 持續學習的重要性
  • 維護微調模型面臨的挑戰
  • 關鍵策略與學習類型（線上、增量、遷移）

數據處理與串流管道

  • 管理不斷變化的數據集
  • 使用小批次與串流API進行線上學習
  • 隨時間變化的數據標籤與註解挑戰

防止災難性遺忘

  • 彈性權重整合（EWC）
  • 重播方法與排練策略
  • 正則化與記憶增強網絡

模型漂移與監控

  • 檢測數據與概念漂移
  • 模型健康與性能衰減的指標
  • 觸發自動模型更新

模型更新的自動化

  • 自動重新訓練與排程策略
  • 與CI/CD和MLOps工作流的整合
  • 管理更新頻率與回滾計劃

持續學習框架與工具

  • Avalanche、Hugging Face Datasets與TorchReplay概覽
  • 平台對持續學習的支持（例如MLflow、Kubeflow）
  • Scala能力與部署考量

真實世界Use Case與架構

  • 基於演化模式的客戶行為預測
  • 工業機器監控與增量改進
  • 變化威脅模型下的詐欺檢測系統

總結與下一步

Sites Published:

United Arab Emirates - Continual Learning and Model Update Strategies for Fine-Tuned Models

Qatar - Continual Learning and Model Update Strategies for Fine-Tuned Models

Egypt - Continual Learning and Model Update Strategies for Fine-Tuned Models

Saudi Arabia - Continual Learning and Model Update Strategies for Fine-Tuned Models

South Africa - Continual Learning and Model Update Strategies for Fine-Tuned Models

Brasil - Continual Learning and Model Update Strategies for Fine-Tuned Models

Canada - Continual Learning and Model Update Strategies for Fine-Tuned Models

中国 - Continual Learning and Model Update Strategies for Fine-Tuned Models

香港 - Continual Learning and Model Update Strategies for Fine-Tuned Models

澳門 - Continual Learning and Model Update Strategies for Fine-Tuned Models

台灣 - Continual Learning and Model Update Strategies for Fine-Tuned Models

USA - Continual Learning and Model Update Strategies for Fine-Tuned Models

Österreich - Continual Learning and Model Update Strategies for Fine-Tuned Models

Schweiz - Continual Learning and Model Update Strategies for Fine-Tuned Models

Deutschland - Continual Learning and Model Update Strategies for Fine-Tuned Models

Czech Republic - Continual Learning and Model Update Strategies for Fine-Tuned Models

Denmark - Continual Learning and Model Update Strategies for Fine-Tuned Models

Estonia - Continual Learning and Model Update Strategies for Fine-Tuned Models

Finland - Continual Learning and Model Update Strategies for Fine-Tuned Models

Greece - Continual Learning and Model Update Strategies for Fine-Tuned Models

Magyarország - Continual Learning and Model Update Strategies for Fine-Tuned Models

Ireland - Continual Learning and Model Update Strategies for Fine-Tuned Models

Luxembourg - Continual Learning and Model Update Strategies for Fine-Tuned Models

Latvia - Continual Learning and Model Update Strategies for Fine-Tuned Models

España - Continual Learning and Model Update Strategies for Fine-Tuned Models

Italia - Continual Learning and Model Update Strategies for Fine-Tuned Models

Lithuania - Continual Learning and Model Update Strategies for Fine-Tuned Models

Nederland - Continual Learning and Model Update Strategies for Fine-Tuned Models

Norway - Continual Learning and Model Update Strategies for Fine-Tuned Models

Portugal - Continual Learning and Model Update Strategies for Fine-Tuned Models

România - Continual Learning and Model Update Strategies for Fine-Tuned Models

Sverige - Continual Learning and Model Update Strategies for Fine-Tuned Models

Türkiye - Continual Learning and Model Update Strategies for Fine-Tuned Models

Malta - Continual Learning and Model Update Strategies for Fine-Tuned Models

Belgique - Continual Learning and Model Update Strategies for Fine-Tuned Models

France - Continual Learning and Model Update Strategies for Fine-Tuned Models

日本 - Continual Learning and Model Update Strategies for Fine-Tuned Models

Australia - Continual Learning and Model Update Strategies for Fine-Tuned Models

Malaysia - Continual Learning and Model Update Strategies for Fine-Tuned Models

New Zealand - Continual Learning and Model Update Strategies for Fine-Tuned Models

Philippines - Continual Learning and Model Update Strategies for Fine-Tuned Models

Singapore - Continual Learning and Model Update Strategies for Fine-Tuned Models

Thailand - Continual Learning and Model Update Strategies for Fine-Tuned Models

Vietnam - Continual Learning and Model Update Strategies for Fine-Tuned Models

India - Continual Learning and Model Update Strategies for Fine-Tuned Models

Argentina - Continual Learning and Model Update Strategies for Fine-Tuned Models

Chile - Continual Learning and Model Update Strategies for Fine-Tuned Models

Costa Rica - Continual Learning and Model Update Strategies for Fine-Tuned Models

Ecuador - Continual Learning and Model Update Strategies for Fine-Tuned Models

Guatemala - Continual Learning and Model Update Strategies for Fine-Tuned Models

Colombia - Continual Learning and Model Update Strategies for Fine-Tuned Models

México - Continual Learning and Model Update Strategies for Fine-Tuned Models

Panama - Continual Learning and Model Update Strategies for Fine-Tuned Models

Peru - Continual Learning and Model Update Strategies for Fine-Tuned Models

Uruguay - Continual Learning and Model Update Strategies for Fine-Tuned Models

Venezuela - Continual Learning and Model Update Strategies for Fine-Tuned Models

Polska - Continual Learning and Model Update Strategies for Fine-Tuned Models

United Kingdom - Continual Learning and Model Update Strategies for Fine-Tuned Models

South Korea - Continual Learning and Model Update Strategies for Fine-Tuned Models

Pakistan - Continual Learning and Model Update Strategies for Fine-Tuned Models

Sri Lanka - Continual Learning and Model Update Strategies for Fine-Tuned Models

Bulgaria - Continual Learning and Model Update Strategies for Fine-Tuned Models

Bolivia - Continual Learning and Model Update Strategies for Fine-Tuned Models

Indonesia - Continual Learning and Model Update Strategies for Fine-Tuned Models

Kazakhstan - Continual Learning and Model Update Strategies for Fine-Tuned Models

Moldova - Continual Learning and Model Update Strategies for Fine-Tuned Models

Morocco - Continual Learning and Model Update Strategies for Fine-Tuned Models

Tunisia - Continual Learning and Model Update Strategies for Fine-Tuned Models

Kuwait - Continual Learning and Model Update Strategies for Fine-Tuned Models

Oman - Continual Learning and Model Update Strategies for Fine-Tuned Models

Slovakia - Continual Learning and Model Update Strategies for Fine-Tuned Models

Kenya - Continual Learning and Model Update Strategies for Fine-Tuned Models

Nigeria - Continual Learning and Model Update Strategies for Fine-Tuned Models

Botswana - Continual Learning and Model Update Strategies for Fine-Tuned Models

Slovenia - Continual Learning and Model Update Strategies for Fine-Tuned Models

Croatia - Continual Learning and Model Update Strategies for Fine-Tuned Models

Serbia - Continual Learning and Model Update Strategies for Fine-Tuned Models

Bhutan - Continual Learning and Model Update Strategies for Fine-Tuned Models

Nepal - Continual Learning and Model Update Strategies for Fine-Tuned Models

Uzbekistan - Continual Learning and Model Update Strategies for Fine-Tuned Models