A two-phase task scheduling framework combining interval Type-2 fuzzy inference with offline proximal policy optimization achieved the best mean performance across five primary operational objectives in fog-cloud computing environments. In multi-workload evaluations, the proposed method, designated as AIT2-FRL, surpassed four existing baselines in makespan, energy consumption, carbon emissions, unreliability, and load imbalance. These performance advantages were supported by statistical significance across all evaluated workload sizes, with the sole exception of the unreliability metric at the smallest evaluated workload.
The study addresses the operational challenges of fog-cloud systems that support latency-sensitive Internet of Things applications. Scheduling independent tasks in these distributed environments represents a multi-objective problem driven by heterogeneous and time-varying computing resources. Distributed nodes differ considerably in processing speed, power demand, network delay, failure tendency, and carbon intensity, while incoming tasks vary in their urgency and resource demands. Existing scheduling heuristics encode only limited operational trade-offs, population-based optimizers frequently demand repeated search routines, and direct reinforcement learning models can obscure decision transparency during deployment.
To overcome these trade-offs, the AIT2-FRL architecture separates scheduling into fuzzy inference and reinforcement learning components. An interval Type-2 fuzzy logic layer captures operational uncertainty regarding task urgency, network congestion, and carbon conditions. This fuzzy layer establishes a one-time sequence for independent tasks and computes context-dependent weighting parameters across makespan, energy, carbon emissions, unreliability, and load imbalance. Proximal policy optimization is utilized exclusively during offline training to calibrate the system’s membership functions and weighting parameters. Once calibrated, a validation-selected checkpoint is frozen for deployment, enabling candidate nodes to be ranked through a deterministic weighted scoring function with linear complexity relative to the number of available nodes.
The authors evaluated the framework across six workload sizes, 17 matched holdout seeds, and resource pools scaling up to 256 nodes governed by common traces and constraints. The experimental design included four comparative baselines, ablation studies, sensitivity analyses, and convergence testing. Although AIT2-FRL demonstrated superior mean values for all primary objectives, secondary assessments showed that these performance gains did not extend uniformly across every operational metric, with specific limitations identified in runtime and secondary response-time measurements.
Subject of Research: Multi-objective task scheduling in fog-cloud computing systems
Article Title: AIT2-FRL: offline PPO-optimized interval type-2 fuzzy scheduling for fair and sustainable fog–cloud systems
Article References: Moazzami, S., Mirzaei, A., Aminian, M., Karimi, R., & Mikaeilvand, N. (2026). AIT2-FRL: offline PPO-optimized interval type-2 fuzzy scheduling for fair and sustainable fog–cloud systems. Scientific Reports. https://doi.org/10.1038/s41598-026-77705-0
Image Credits: AI Generated
DOI: 10.1038/s41598-026-77705-0
Keywords: fog-cloud systems, task scheduling, interval type-2 fuzzy inference, proximal policy optimization, makespan, energy consumption, carbon emissions, load imbalance, unreliability, internet of things, latency-sensitive applications, offline reinforcement learning, heterogeneous computing, membership functions, deterministic ranking, holdout validation, multi-objective optimization
Cite Scienmag News
Josephine Dean. (November 12, 2026). Offline Optimized Fuzzy Scheduler Improves Five Primary Objectives in Fog-Cloud Systems. Scienmag. https://scienmag.com/offline-optimized-fuzzy-scheduler-improves-five-primary-objectives-in-fog-cloud-systems/
Josephine Dean. "Offline Optimized Fuzzy Scheduler Improves Five Primary Objectives in Fog-Cloud Systems." Scienmag, 12 November 2026, https://scienmag.com/offline-optimized-fuzzy-scheduler-improves-five-primary-objectives-in-fog-cloud-systems/. Accessed 21 November 2026.
Josephine Dean. "Offline Optimized Fuzzy Scheduler Improves Five Primary Objectives in Fog-Cloud Systems." Scienmag. November 12, 2026. https://scienmag.com/offline-optimized-fuzzy-scheduler-improves-five-primary-objectives-in-fog-cloud-systems/








