Explore privacy-preserving algorithms for decentralized collaborative learning in this comprehensive lecture by Dr Aurélien Bellet from Inria. Delve into key principles of gossip algorithms, personalized learning, and model propagation in distributed settings. Examine the convergence results for asynchronous gossip algorithms and their application in broadcast settings. Investigate the formulation of collaborative learning problems and the implementation of differential privacy to ensure data protection. Gain insights into large-scale machine learning, distributed algorithms, and privacy-preserving techniques applicable to various domains including NLP, speech recognition, and computer vision.
Privacy-Preserving Algorithms for Decentralised Collaborative Learning - Dr Aurélien Bellet
Alan Turing Institute via YouTube
Overview
Syllabus
LEARNING FROM CONNECTED DEVICES DATA
EXTREME APPROACH 1: CENTRALIZED LEARNING
OUR APPROACH: FULLY DECENTRALIZED LEARNING
KEY PRINCIPLES GOSSIP ALGORITHM
THIS WORK: PERSONALIZED LEARNING
PROBLEM SETTING
MODEL PROPAGATION: PROBLEM FORMULATION
ASYNCHRONOUS GOSSIP ALGORITHM
CONVERGENCE RESULT
ALGORITHM IN THE BROADCAST SETTING
CONVERGENCE IN BROADCAST SETTING
COLLABORATIVE LEARNING PROBLEM FORMULATION
DIFFERENTIAL PRIVACY
PRIVACY GUARANTEE
Taught by
Alan Turing Institute
