Interview with Blaise Aguéra y Arcas
summary
Argues that federated learning separates the learning process from data sharing and becomes practical because large models enable rapid fine‑tuning, prompting and one‑shot adaptation after pretraining, not by scaling classic FL algorithms alone. Notes differential privacy is mathematically precise while actual privacy is psychological and subjective. Warns that treating models as proxies for companies risks a surveillance state as on‑device augmentation spreads, giving the Lèse‑majesté example as a concrete legal privacy failure to avoid.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 1:51What advantages does federated learning provide compared to other privacy preserving techniques?
- 3:54Would you anticipate that federated learning will play a bigger role in policy initiatives?
- 4:39Why are conferences such as the Privacy Symposium important?
- 5:11What is your priority objective and vision for the year to come?
- 6:35Any messages you would like to share?
related talks
29mBalancing Thin Line Between Data Intelligence & Privacy
31mFederated Learning: So Close and Yet So Out of Reach
40mDemystifying Decentralized Fair Models for Federated Machine Learning
28mA Privacy Enhancing Federated Learning Framework for Multi-Party Distributed ..
01h 20mUnlocking societal benefits of AI with privacy protective technologies.
44mPrivacy by Design Machine Learning
15mLearning and Unlearning Your Data in Federated Settings
34mDecentralized Machine Learning