Distributed Optimization and Privacy Preservation with Quantized Communication

Talk by Dr. Apostolos I. Rikos

March 17, 2023 Friday

Abstract:

The problem of distributed optimization has received extensive attention over recent years from the control and machine learning communities, due to its wide area of applications. However, for large-scale networks with possibly limited bandwidth capacity, communication overhead becomes a major bottleneck. Quantization of information is one of the major approaches to overcome this issue.

In our presentation we will focus on an unconstrained distributed optimization problem. We will present a distributed optimization algorithm which relies on gradient descent, and is combined with a finite time coordination algorithm where nodes are exchanging quantized messages. We will show that our proposed algorithm linearly converges to a neighborhood of the optimal solution, determined by the quantization level. Then, we will present a privacy preservation protocol which relies on quantized multiple state decomposition and could be combined with the aforementioned optimization algorithm. We will present the topological conditions according to which our protocol guarantees privacy preservation. Then, we will show that our privacy protocol converges in finite time while guaranteeing quantized communication between nodes. Finally, we will conclude our presentation with simulation results over various networks. 

Speaker Bio:

Dr. Apostolos I. Rikos

Postdoctoral Researcher, Division of Decision and Control Systems, KTH Royal Institute of Technology

I am a postdoctoral researcher at KTH Royal Institute of Technology. I am working with Prof. Karl Henrik Johansson on distributed network control, distributed optimization, privacy preservation, and unsupervised learning. I received the B.Sc., M.Sc and Ph.D. degrees in Electrical Engineering from the Department of Electrical and Computer Engineering, University of Cyprus in 2010, 2012 and 2018 respectively. My supervisor was Prof. Christoforos N. Hadjicostis. My Ph.D. thesis focused on distributed control and coordination of multi agent networks, entitled ”Distributed Weight Balancing in Directed Topologies”. From November 2018 until January 2020, I joined KIOS Research and Innovation Center of Excellence in Cyprus as a Special Scientist and Associate Lecturer. Since February 2020, I have been a postdoctoral researcher at KTH Royal Institute of Technology.