Federated and distributed inference leveraging sensing and communication in the computing continuum

Introduction

The integration of sensing and communication is attracting fervent research activity and will result in a myriad of contextual data that, if properly processed, may enable a better understanding of local and global phenomena while increasing the quality, security, and efficiency of our ecosystems.

Computing Continuum

The computing continuum offers a timely and unique solution for processing such a massive volume of sensed data, as it provides virtually unlimited and widely distributed computing resources.

Challenges in Data Analysis

Nevertheless, the deployment of data analysis at the edge or in the cloud has many implications regarding:

• Latency
• Privacy
• Security
• Data integrity

As we learn how to sense ubiquitously and build tools to handle the sensed data, the greatest challenge is to understand how and where to process them.

Project Purpose

The purpose of this project is the development of a pioneering framework to guide the design of federated and distributed inference systems, leveraging sensing and communication and harnessing the computing continuum.

Framework Components

The framework will build on:

1. The definition of statistical and mathematical models for the sensed data, which capture the complex and interrelated phenomena underpinning sensing and communication systems, with different levels of integration.
2. The development of cloud-native inference algorithms, mainly distributed and parallelized, with scalable complexity that can be adapted to dynamic performance requirements.
3. The design of orchestration strategies to guide the flexible deployment of the inference process at the edge and in the cloud with a dynamic allocation of the computing resources.

Aim of the Project

The aim is to overcome the paradigmatic accuracy-complexity trade-off that has driven distributed inference for decades, leading to a paradigm shift that encompasses multi-level performance indicators beyond accuracy, including:

• Latency
• Integrity
• Privacy
• Security aspects

Additionally, the project will explore how these factors impact the confidence in the inferred phenomena.