Distributed Machine Learning over Unreliable Networks

Online learning is a variety of sequential decision-making problems under uncertainty, envisaged by an agent interacting with an unknown environment by successively selecting an action from a set of available actions. The environment can have different states. Each action, if it is selected, returns some state-dependent reward. Often, the goal of the decision-maker is to satisfy some optimality condition over the interaction horizon, which is often defined in terms of accumulated discounted reward or the accumulated regret. Graph signal processing (GSP) is a recently-established branch of signal processing. The goal of GSP is to handle the excessive amount of data that is collected in various circumstances, for example, from a sensor network, in a fast and efficient manner.

The mutual relation between GSP and machine learning is an emerging and promising field of research. In this project, the goal is to develop, adapt, and integrate the tools from graph signal processing to enable fast and efficient online learning and decision-making under uncertainty. The link between graph signal processing and online learning can be established by creative and appropriate modeling.

This project is financed by the German research foundation (DFG) in the framework of excellence cluster Machine Learning: New Perspectives for Science. The duration is 01.2021-12.2023. The responsible researcher for this project is Mr. Behzad Nourani Koliji.