- 了解機器學習工作流程和模型訓練
- 具備Python和常見ML框架（如PyTorch或TensorFlow）的經驗
- 熟悉基本安全或威脅建模概念會有所幫助
目標受眾
- 機器學習工程師
- 網絡安全分析師
- AI研究人員和模型驗證團隊
保护AI模型是一门防御机器学习系统免受特定威胁的学科，这些威胁包括对抗性输入、数据投毒、反转攻击和隐私泄露。
本课程为讲师引导的培训（线上或线下），面向中级机器学习与网络安全专业人士，旨在帮助他们理解并缓解针对AI模型的新兴威胁，使用概念框架与实操防御手段，如鲁棒训练与差分隐私。
通过本培训，参与者将能够：
- 识别并分类AI特定威胁，如对抗性攻击、反转和投毒。
- 使用Adversarial Robustness Toolbox (ART)等工具模拟攻击并测试模型。
- 应用实际防御手段，包括对抗性训练、噪声注入和隐私保护技术。
- 在生产环境中设计威胁感知的模型评估策略。
课程形式
- 互动讲座与讨论。
- 大量练习与实践。
- 在实时实验室环境中进行实操实施。
课程定制选项
- 如需为本课程定制培训，请联系我们安排。
AI威胁建模简介
- AI系统为何容易受到攻击？
- AI攻击面与传统系统的对比
- 关键攻击向量：数据、模型、输出和接口层
针对AI模型的对抗攻击
- 理解对抗样本和扰动技术
- 白盒攻击与黑盒攻击
- FGSM、PGD和DeepFool方法
- 可视化和制作对抗样本
模型反演与隐私泄露
- 从模型输出推断训练数据
- 成员推断攻击
- 分类和生成模型中的隐私风险
数据投毒与后门注入
- 投毒数据如何影响模型行为
- 基于触发的后门和木马攻击
- 检测与净化策略
鲁棒性与防御技术
- 对抗训练与数据增强
- 梯度掩码与输入预处理
- 模型平滑与正则化技术
隐私保护AI防御
- 差分隐私简介
- 噪声注入与隐私预算
- 联邦学习与安全聚合
AI Security实践
- 威胁感知的模型评估与部署
- 在应用场景中使用ART（对抗鲁棒性工具箱）
- 行业案例研究：真实世界的数据泄露与缓解措施
总结与下一步
United Arab Emirates - Securing AI Models: Threats, Attacks, and Defenses
Qatar - Securing AI Models: Threats, Attacks, and Defenses
Egypt - Securing AI Models: Threats, Attacks, and Defenses
Saudi Arabia - Securing AI Models: Threats, Attacks, and Defenses
South Africa - Securing AI Models: Threats, Attacks, and Defenses
Brasil - Securing AI Models: Threats, Attacks, and Defenses
Canada - Securing AI Models: Threats, Attacks, and Defenses
中国 - Securing AI Models: Threats, Attacks, and Defenses
香港 - Securing AI Models: Threats, Attacks, and Defenses
澳門 - Securing AI Models: Threats, Attacks, and Defenses
台灣 - Securing AI Models: Threats, Attacks, and Defenses
USA - Securing AI Models: Threats, Attacks, and Defenses
Österreich - Securing AI Models: Threats, Attacks, and Defenses
Schweiz - Securing AI Models: Threats, Attacks, and Defenses
Deutschland - Securing AI Models: Threats, Attacks, and Defenses
Czech Republic - Securing AI Models: Threats, Attacks, and Defenses
Denmark - Securing AI Models: Threats, Attacks, and Defenses
Estonia - Securing AI Models: Threats, Attacks, and Defenses
Finland - Securing AI Models: Threats, Attacks, and Defenses
Greece - Securing AI Models: Threats, Attacks, and Defenses
Magyarország - Securing AI Models: Threats, Attacks, and Defenses
Ireland - Securing AI Models: Threats, Attacks, and Defenses
Luxembourg - Securing AI Models: Threats, Attacks, and Defenses
Latvia - Securing AI Models: Threats, Attacks, and Defenses
España - Securing AI Models: Threats, Attacks, and Defenses
Italia - Securing AI Models: Threats, Attacks, and Defenses
Lithuania - Securing AI Models: Threats, Attacks, and Defenses
Nederland - Securing AI Models: Threats, Attacks, and Defenses
Norway - Securing AI Models: Threats, Attacks, and Defenses
Portugal - Securing AI Models: Threats, Attacks, and Defenses
România - Securing AI Models: Threats, Attacks, and Defenses
Sverige - Securing AI Models: Threats, Attacks, and Defenses
Türkiye - Securing AI Models: Threats, Attacks, and Defenses
Malta - Securing AI Models: Threats, Attacks, and Defenses
Belgique - Securing AI Models: Threats, Attacks, and Defenses
France - Securing AI Models: Threats, Attacks, and Defenses
日本 - Securing AI Models: Threats, Attacks, and Defenses
Australia - Securing AI Models: Threats, Attacks, and Defenses
Malaysia - Securing AI Models: Threats, Attacks, and Defenses
New Zealand - Securing AI Models: Threats, Attacks, and Defenses
Philippines - Securing AI Models: Threats, Attacks, and Defenses
Singapore - Securing AI Models: Threats, Attacks, and Defenses
Thailand - Securing AI Models: Threats, Attacks, and Defenses
Vietnam - Securing AI Models: Threats, Attacks, and Defenses
India - Securing AI Models: Threats, Attacks, and Defenses
Argentina - Securing AI Models: Threats, Attacks, and Defenses
Chile - Securing AI Models: Threats, Attacks, and Defenses
Costa Rica - Securing AI Models: Threats, Attacks, and Defenses
Ecuador - Securing AI Models: Threats, Attacks, and Defenses
Guatemala - Securing AI Models: Threats, Attacks, and Defenses
Colombia - Securing AI Models: Threats, Attacks, and Defenses
México - Securing AI Models: Threats, Attacks, and Defenses
Panama - Securing AI Models: Threats, Attacks, and Defenses
Peru - Securing AI Models: Threats, Attacks, and Defenses
Uruguay - Securing AI Models: Threats, Attacks, and Defenses
Venezuela - Securing AI Models: Threats, Attacks, and Defenses
Polska - Securing AI Models: Threats, Attacks, and Defenses
United Kingdom - Securing AI Models: Threats, Attacks, and Defenses
South Korea - Securing AI Models: Threats, Attacks, and Defenses
Pakistan - Securing AI Models: Threats, Attacks, and Defenses
Sri Lanka - Securing AI Models: Threats, Attacks, and Defenses
Bulgaria - Securing AI Models: Threats, Attacks, and Defenses
Bolivia - Securing AI Models: Threats, Attacks, and Defenses
Indonesia - Securing AI Models: Threats, Attacks, and Defenses
Kazakhstan - Securing AI Models: Threats, Attacks, and Defenses
Moldova - Securing AI Models: Threats, Attacks, and Defenses
Morocco - Securing AI Models: Threats, Attacks, and Defenses
Tunisia - Securing AI Models: Threats, Attacks, and Defenses
Kuwait - Securing AI Models: Threats, Attacks, and Defenses
Oman - Securing AI Models: Threats, Attacks, and Defenses
Slovakia - Securing AI Models: Threats, Attacks, and Defenses
Kenya - Securing AI Models: Threats, Attacks, and Defenses
Nigeria - Securing AI Models: Threats, Attacks, and Defenses
Botswana - Securing AI Models: Threats, Attacks, and Defenses
Slovenia - Securing AI Models: Threats, Attacks, and Defenses
Croatia - Securing AI Models: Threats, Attacks, and Defenses
Serbia - Securing AI Models: Threats, Attacks, and Defenses
Bhutan - Securing AI Models: Threats, Attacks, and Defenses
Nepal - Securing AI Models: Threats, Attacks, and Defenses
Uzbekistan - Securing AI Models: Threats, Attacks, and Defenses