Ever wondered what happens to your data after you delete your account? While the database might forget you, the AI models trained on your data still "remember" your patterns. This talk introduces machine unlearning - the fascinating field of teaching AI to truly forget.
In this beginner-friendly session, you'll discover why machine unlearning matters more than ever. From GDPR's "right to be forgotten" to removing bias from hiring algorithms, the ability to make AI forget specific information is crucial for privacy, fairness, and safety.
I'll demonstrate some practical and theoretical Python approaches you can implement today:
- SISA Framework - Split training data into shards, making forgetting 5x faster than retraining fromscratch
- Gradient Ascent - Teach models to be confidently wrong about specific data
- Strategic Noise Injection - Add controlled noise to confuse AI about unwanted memories
I'll explain the key concepts and walk through Python examples, showing how different unlearning approaches work in practice. While some advanced techniques exist, I'll focus on methods that are approachable for those getting started with machine learning.
Whether you're building your first ML model or working on production systems, understanding machine unlearning helps you create more ethical and privacy-conscious AI. You'll leave with practical code examples, resources for further learning, and a new perspective on responsible AI development.
You just need curiosity about making AI more trustworthy and accountable.
