Cloud computing is extremely important for 5G/6G wireless networks as it allows to solve many of the complex communication and machine learning problems in a distributed fashion using software, rather than costly, time-intensive, and often slow-to-evolve hardware. The evolution of telecom infrastructures toward 6G assumes moving computation from the central cloud to edge computing resources that are included in each node of the network. Edge computing vastly reduces the amount of data to be sent in the network and thus the network load. It also reduces the latency. At the same time there are several problems connected to the decentralized nature of edge computing. Indeed, the central node can receive the results from processing nodes with delay or do not receive part of the results at all. Such nodes are called straggler nodes. Another important aspect is privacy, i.e., private user information should not be disclosed while processing and communicating. To meet 6G requirements new fast, reliable, and secure distributed computing methods should be developed.

We have been working on a very close topic – reliable and secure distributed storage, my group has published more than 10 publications on this topic. I would especially note publications [1-3] in IEEE Transaction on Information Theory (impact-factor 2.501, Q1) and IEEE Transactions on Information Forensics & Security (impact-factor 7.178, Q1). We plan to carry out research in the following main directions:

1. Coded distributed computing algorithms. A severe problem in distributed computation are stragglers [4] as waiting for their response causes a tremendous delay for the machine learning algorithm. Therefore, erasure-correcting codes are planned to be utilized with the goal to be able to reconstruct the overall result from many (but not all) workers.

2. Private distributed computing. We plan to focus on learning tasks, this line of research is called federated learning [5]. Our initial goal is to derive fundamental trade-offs in between privacy, communication, and accuracy, as well as investigate how privacy constraints affect the convergence of ML algorithms. Finally, we are going to propose private distributed computing based on a local differential privacy approach.