Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Active Inference Agents for Heterogeneous and Lifelong Federated Learning

Anastasiya Danilenka, Alireza Furutanpey, Victor Casamayor Pujol, Boris Sedlak, Anna Lackinger, Maria Ganzha, Marcin Paprzycki, Schahram Dustdar

TL;DR

This work tackles lifelong heterogeneous federated learning under non-IID data and dynamic resource constraints by introducing adaptive active inference (AIF) agents that operate with high-level SLO targets. The method builds a discrete Bayesian Network world model to capture causal factors affecting SLO fulfillment and uses the expected free energy ($EFE = - PV - IG$) to select training configurations, balancing Pragmatic Value ($PV$) and Information Gain ($IG$). The authors provide a rigorous experimental evaluation on a heterogeneous physical testbed, showing that AIF agents can autonomously adapt configurations to environmental changes and recover from concept and quantity drifts, achieving up to 98% SLO fulfillment. The results indicate that AIF offers a practical, continuous adaptation mechanism for pervasive FL systems and can outperform static baselines and standard hyperparameter optimization in dynamic settings.

Abstract

Handling heterogeneity and unpredictability are two core problems in pervasive computing. The challenge is to seamlessly integrate devices with varying computational resources in a dynamic environment to form a cohesive system that can fulfill the needs of all participants. Existing work on adaptive systems typically focuses on optimizing individual variables or low-level Service Level Objectives (SLOs), such as constraining the usage of specific resources. While low-level control mechanisms permit fine-grained control over a system, they introduce considerable complexity, particularly in dynamic environments. To this end, we propose drawing from Active Inference (AIF), a neuroscientific framework for designing adaptive agents. Specifically, we introduce a conceptual agent for heterogeneous pervasive systems that permits setting global systems constraints as high-level SLOs. Instead of manually setting low-level SLOs, the system finds an equilibrium that can adapt to environmental changes. We demonstrate the viability of our AIF agents with an extensive experiment design, using heterogeneous and lifelong federated learning as an application scenario. We conduct our experiments on a physical testbed of devices with different resource types and vendor specifications. The results provide convincing evidence that an AIF agent can adapt a system to environmental changes. In particular, the AIF agent can balance competing SLOs in resource heterogeneous environments to ensure up to 98% fulfillment rate.

Adaptive Active Inference Agents for Heterogeneous and Lifelong Federated Learning

TL;DR

This work tackles lifelong heterogeneous federated learning under non-IID data and dynamic resource constraints by introducing adaptive active inference (AIF) agents that operate with high-level SLO targets. The method builds a discrete Bayesian Network world model to capture causal factors affecting SLO fulfillment and uses the expected free energy () to select training configurations, balancing Pragmatic Value () and Information Gain (). The authors provide a rigorous experimental evaluation on a heterogeneous physical testbed, showing that AIF agents can autonomously adapt configurations to environmental changes and recover from concept and quantity drifts, achieving up to 98% SLO fulfillment. The results indicate that AIF offers a practical, continuous adaptation mechanism for pervasive FL systems and can outperform static baselines and standard hyperparameter optimization in dynamic settings.

Abstract

Handling heterogeneity and unpredictability are two core problems in pervasive computing. The challenge is to seamlessly integrate devices with varying computational resources in a dynamic environment to form a cohesive system that can fulfill the needs of all participants. Existing work on adaptive systems typically focuses on optimizing individual variables or low-level Service Level Objectives (SLOs), such as constraining the usage of specific resources. While low-level control mechanisms permit fine-grained control over a system, they introduce considerable complexity, particularly in dynamic environments. To this end, we propose drawing from Active Inference (AIF), a neuroscientific framework for designing adaptive agents. Specifically, we introduce a conceptual agent for heterogeneous pervasive systems that permits setting global systems constraints as high-level SLOs. Instead of manually setting low-level SLOs, the system finds an equilibrium that can adapt to environmental changes. We demonstrate the viability of our AIF agents with an extensive experiment design, using heterogeneous and lifelong federated learning as an application scenario. We conduct our experiments on a physical testbed of devices with different resource types and vendor specifications. The results provide convincing evidence that an AIF agent can adapt a system to environmental changes. In particular, the AIF agent can balance competing SLOs in resource heterogeneous environments to ensure up to 98% fulfillment rate.

Paper Structure

This paper contains 24 sections, 7 equations, 17 figures, 2 tables, 1 algorithm.

Table of Contents

  1. introduction
  2. Background and Related Work
  3. Lifelong Heterogeneous Federated Learning
  4. Lifelong Federated Learning
  5. Heterogeneous Federated Learning
  6. Service Level Objectives for Federated Learning
  7. Active Inference
  8. Problem Statement
  9. Proposed Method
  10. Learning a Simple World Model
  11. EFE and SLO-aware Configuration Selection
  12. Evaluation
  13. Experiment Design
  14. Test Bed
  15. Implementation Details
  16. ...and 9 more sections

Figures (17)

  • Figure 1: Heterogeneous FL with data streams and AIF agents
  • Figure 2: Sequence diagram for one FL round of the proposed method
  • Figure 3: BN structure update throughout FL training rounds (blue vertices represent SLOs, green -- configuration variables)
  • Figure 4: Mean cumulative SLOs fulfillment with two quantity drifts (red lines mark the drift start). Semitransparent lines show mean SLO fulfillment at a single federated round.
  • Figure 5: Mean EFE with two quantity drifts with each line representing one possible training configuration. Lower values represent configurations that the AIF agents favor.
  • ...and 12 more figures