Decoding Machine UNLEARNING: Can AI Truly Forget? | Audio | Mono | PODCAST | AI-Papers
Автор: Manohar | Hey World
Загружено: 2025-05-14
Просмотров: 74
Описание:
Can Machines Truly Forget? 🔍 Exploring the Limits of Machine Unlearning
In this episode, we dive into the groundbreaking research paper “Mirror Mirror on the Wall, Have I Forgotten it All?” by Brimhall, Mathew, Fendley, Cao, and Green (2025) (URL: https://arxiv.org/abs/2505.08138). The authors introduce computational unlearning, a powerful new framework to test whether AI models can truly forget data—like it was never there.
We break down:
What machine unlearning is and why it matters
Why retraining isn’t enough
How current unlearning methods fall short
The role of randomness, cryptography, and differential privacy
The trade-off between forgetting and performance
📉 From failed defenses to 100% successful attacks, this paper sets a bold new standard for privacy in machine learning… and shows just how far we still have to go.
🎙️ Perfect for listeners into AI ethics, privacy, model training, and the hidden challenges behind data deletion.
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
