感謝您提交詢問！我們的一位團隊成員將在短時間內與您聯繫。
感謝您提交預訂！我們的一位團隊成員將在短時間內與您聯繫。
課程簡介
Foundations of Safe and Fair AI
- Core concepts: safety, bias, fairness, and transparency.
- Types of bias: dataset, representation, and algorithmic.
- Overview of regulatory frameworks (e.g., EU AI Act, GDPR).
Bias in Fine-Tuned Models
- How fine-tuning can introduce or amplify bias.
- Case studies and real-world failures.
- Identifying bias in datasets and model predictions.
Techniques for Bias Mitigation
- Data-level strategies (rebalancing, augmentation).
- In-training strategies (regularization, adversarial debiasing).
- Post-processing strategies (output filtering, calibration).
Model Safety and Robustness
- Detecting unsafe or harmful outputs.
- Handling adversarial inputs.
- Red teaming and stress testing fine-tuned models.
Auditing and Monitoring AI Systems
- Bias and fairness evaluation metrics (e.g., demographic parity).
- Explainability tools and transparency frameworks.
- Ongoing monitoring and governance practices.
Toolkits and Hands-On Practice
- Using open-source libraries (e.g., Fairlearn, Transformers, CheckList).
- Hands-on: Detecting and mitigating bias in a fine-tuned model.
- Generating safe outputs through prompt design and constraints.
Enterprise Use Cases and Compliance Readiness
- Best practices for integrating safety into LLM workflows.
- Documentation and model cards for compliance.
- Preparing for audits and external reviews.
Summary and Next Steps
最低要求
- A foundational understanding of machine learning models and training processes.
- Practical experience with fine-tuning and Large Language Models (LLMs).
- Familiarity with Python and Natural Language Processing (NLP) concepts.
Target Audience
- AI compliance teams.
- Machine learning engineers.
14 小時
