Pareto-Optimal Bit Allocation for Collaborative Intelligence

TL;DR

This paper develops a Pareto-optimal bit allocation framework for collaborative AI systems split between edge and cloud, optimizing feature coding to improve multi-task performance.

Contribution

It introduces a convex distortion-rate model for feature coding, providing closed-form solutions and Pareto set characterizations for multi-stream, multi-task systems.

Findings

Closed-form bit allocation solutions for single-task systems

Analytical Pareto set characterization for 2-stream, multi-task systems

Bounds on Pareto set for 3-stream, multi-task systems

Abstract

In recent studies, collaborative intelligence (CI) has emerged as a promising framework for deployment of Artificial Intelligence (AI)-based services on mobile/edge devices. In CI, the AI model (a deep neural network) is split between the edge and the cloud, and intermediate features are sent from the edge sub-model to the cloud sub-model. In this paper, we study bit allocation for feature coding in multi-stream CI systems. We model task distortion as a function of rate using convex surfaces similar to those found in distortion-rate theory. Using such models, we are able to provide closed-form bit allocation solutions for single-task systems and scalarized multi-task systems. Moreover, we provide analytical characterization of the full Pareto set for 2-stream k-task systems, and bounds on the Pareto set for 3-stream 2-task systems. Analytical results are examined on a variety of DNN models from the literature to demonstrate wide applicability of the results