Pre-Deployment Complexity Estimation for Federated Perception Systems

Abstract

Edge AI systems increasingly rely on federated learning to train perception models in distributed, privacy-preserving, and resource-constrained environments. Yet, before training begins, practitioners often lack practical tools to estimate how difficult a federated learning task will be in terms of achievable accuracy and communication cost. This paper presents a classifier-agnostic, pre-deployment framework for estimating learning complexity in federated perception systems by jointly modeling intrinsic properties of the data and characteristics of the distributed environment. The proposed complexity metric integrates dataset attributes such as dimensionality, sparsity, and heterogeneity with factors related to the composition of participating clients. Using federated learning as a representative distributed training setting, we examine how learning difficulty varies across different federated configurations. Experiments on multiple variants of the MNIST dataset and CIFAR dataset show that the proposed metric strongly correlates with federated learning performance and the communication effort required to reach fixed accuracy targets. These findings suggest that complexity estimation can serve as a practical diagnostic tool for resource planning, dataset assessment, and feasibility evaluation in edge-deployed perception systems.

Figures (5)

• Figure 1: Illustration of federated learning paths across communication rounds. Each node represents an intermediate training state obtained from a subset of participating clients. Different paths correspond to distinct federated configurations, each associated with potentially different learning difficulty.
• Figure 2: (a) Shallow CNN accuracy vs. benchmark accuracy for Fashion-MNIST, Handwritten-MNIST and EMNIST-Digits, (b) Federated environment complexity $f(d)$ vs. shallow federated learning accuracy for Handwritten-MNIST when $f(X)$ is fixed, (c) Federated environment complexity $f(d)$ vs. shallow federated learning accuracy for Fashion-MNIST when $f(X)$ is fixed.
• Figure 3: (a) Federated environment complexity $f(d)$ vs. effort (communication rounds) for MNIST and Fashion-MNIST, (b) Federated accuracy vs. Federated intrinsic function $f(X)$, reflecting heterogeneity (entropy), for Fashion-MNIST, Handwritten-MNIST and EMNIST-Digits, with $f(d)$ fixed at $2$, (c) Federated accuracy vs. Federated intrinsic function $f(X)$, reflecting sparsity (number of sparse components for explaining $80\%$ of variance), for Fashion-MNIST, Handwritten-MNIST and EMNIST-Digits with $f(d)$ fixed at $2$.
• Figure 4: (a) Federated intrinsic function $f(X)$ (reflecting heterogeneity) vs. effort (communication rounds), (b) Federated complexity $F(d,X)$ vs. shallow Federated learning accuracy, considering maximum accuracy $(R^2=0.81)$.
• Figure 5: Federated complexity $F(d,X)$ vs. shallow Federated learning accuracy, considering average accuracy $(R^2=0.85)$.